By Matt Grobar
Film Reporter
Magnolia Pictures has snapped up U.S. rights to Master Gardener — the latest starry feature from Academy Award nominee Paul Schrader (First Reformed), which world premiered at this year’s Venice Film Festival and most recently screened at the New York Film Festival. Pic is slated for release next year.
The crime thriller scripted by Schrader — marking the last film in his recent trilogy including First Reformed and The Card Counter — follows Narvel Roth (Loving‘s Joel Edgerton), the meticulous horticulturist of Gracewood Gardens. He is as much devoted to tending the grounds of this beautiful and historic estate as to pandering to his employer, the wealthy dowager Mrs. Haverhill (Avatar‘s Sigourney Weaver). When Mrs. Haverhill demands that he take on her wayward and troubled great-niece Maya (Black Adam‘s Quintessa Swindell) as a new apprentice, chaos enters Narvel’s spartan existence, unlocking dark secrets from a buried violent past that threaten them all.    

Master Gardener was produced by Schrader’s manager, David Gonzales, as well as Amanda Crittenden (The Nightingale, HBO’s The Leftovers) and Scott LaStaiti (The Infiltrator, Jane Got a Gun). Ottocento’s Luisa Law, Flickstar’s Jamieson McClurg and KOJO Studios’ Dale Roberts and Linda Ujuk served as exec producers.
“Horticulture has never seemed so fraught with mystery and peril as in MASTER GARDENER,” said Magnolia Pictures President Eamonn Bowles. “Paul Schrader, ably abetted by his remarkable cast, finds fertile soil for exploring his fascinatingly complex themes in this truly exceptional thriller.”
Added Schrader: “Magnolia has been a class act for over twenty years, and I eagerly anticipate and appreciate their distribution of MASTER GARDENER.”
Magnolia EVP Dori Begley and SVP of Acquisitions John Von Thaden negotiated the deal for Master Gardener on behalf of the indie film studio, with David Gonzales at Northern Lights and Gabrielle Stewart at HanWay Films on behalf of the filmmakers.
Subscribe to Deadline Breaking News Alerts and keep your inbox happy.
Signup for Breaking News Alerts & Newsletters
Get our latest storiesin the feed of your favorite networks
We want to hear from you! Send us a tip using our annonymous form.
Sign up for our breaking news alerts
Deadline is a part of Penske Media Corporation. © 2022 Deadline Hollywood, LLC. All Rights Reserved.

source

The highly anticipated deluxe edition reissue of John Mellencamp’s beloved album Scarecrow is out now on Mercury/UMe, featuring a massive collection of bonus tracks, rarities and more never previously shared by Mellencamp. Stream/purchase HERE.
The album features two previously unreleased tracks, “Smart Guys” and “Carolina Shag,” with accompanying lyric videos directed by Shan Dan Horan.
The reissue product suite includes a Super Deluxe edition (2 CDs, 180-gram LP, 7” Single, Blu-ray with ATMOS and Hi-Res Stereo mixes, booklet, poster and lithographs); 2CD & Digital Deluxe edition; 180-gram LP and Deluxe LP. All configurations feature an all-new remix and remaster. The Super Deluxe and 2 CD Deluxe editions include previously unreleased bonus tracks and alternate versions, a booklet full of rare photographs and all-new liner notes by acclaimed author and music critic Anthony DeCurtis. The new ATMOS mix and Digital Deluxe edition (with all bonus tracks) will also be available to stream via DSPs.
The Super Deluxe edition includes a CD of the new stereo mix of the album plus one existing bonus track, a CD of rare and previously unreleased bonus tracks, the ATMOS and hi-res stereo mix of the album and hi-res stereo bonus tracks on Blu-ray audio disc, a 180g LP of the original album with a new stereo mix (½ speed), reproduction of the original “Small Town” picture sleeve 7” vinyl single, as well as a photobook, poster, lithographs and essay.
The LP and LP Deluxe versions feature an all-new stereo mix of the album on 180g vinyl (½ speed). The Deluxe vinyl also includes an exclusive lithograph and deluxe packaging. The 2CD/Digital Deluxe edition includes the newly remixed and remastered album plus previously unreleased bonus tracks.
His eighth studio album, Scarecrow was originally released in 1985 and is one of Mellencamp’s most beloved albums. The album contains three chart topping hits—“R.O.C.K In The U.S.A. (A Salute To 60’s Rock),” “Lonely Ol’ Night” and “Small Town”—and reached #2 on the U.S. chart.
Mellencamp’s massive 76-night tour across North America kicks off in February; the announcement came live from the opening of his new, permanent exhibition at the Rock & Roll Hall of Fame. The Live and In Person 2023 tour begins in Bloomington, Indiana and includes multiple nights in New York City, Chicago, Vancouver, Nashville, Clearwater, Minneapolis and more. See below for a complete list of dates; tickets are on sale to the general public now.

Mellencamp is a member of the Rock and Roll and Songwriters Halls of Fame, a recipient of the John Steinbeck Award, ASCAP Foundation’s Champion Award, The Woody Guthrie Award, the Americana Music Association’s Lifetime Achievement Award, and most recently, the Founders Award, the top honor assigned by the American Society of Composers, Authors and Publishers.
Tickets to the “Live and In Person 2023” tour are available HERE.
 
John Mellencamp—Live and In Person 2023 Tour
February 5—Indiana University Auditorium—Bloomington, IN
February 6— Indiana University Auditorium—Bloomington, IN
February 8—Ovens Auditorium—Charlotte, NC
February 10—Fox Theatre—Atlanta, GA
 February 11—Jacksonville Center for the Arts | Moran Theater—Jacksonville, FL
February 13—Ruth Eckerd Hall—Clearwater, FL
February 14—Ruth Eckerd Hall—Clearwater, FL
February 15—Ruth Eckerd Hall—Clearwater, FL
February 18—Walt Disney Theater at Dr. Phillips Center for the Performing Arts—Orlando, FL
February 19—Walt Disney Theater at Dr. Phillips Center for the Performing Arts—Orlando, FL
February 21—Broward Center for the Performing Arts—Ft. Lauderdale, FL
February 22—Broward Center for the Performing Arts—Ft. Lauderdale, FL
February 24—Saenger Theatre—New Orleans, LA
February 25—Smart Financial Centre—Sugar Land, TX
February 26—Majestic Theatre—San Antonio, TX
February 28—Bass Concert Hall—Austin, TX
March 1—Texas Trust CU Theatre at Grande Prairie—Dallas, TX
March 11—Keller Auditorium—Portland, OR
March 13—Orpheum—Vancouver, BC
March 14—Orpheum—Vancouver, BC
March 15—Paramount Theatre—Seattle, WA
March 17—Golden Gate Theatre—San Francisco, CA
March 18—Golden Gate Theatre—San Francisco, CA
March 19—Saroyan Theatre—Fresno, CA
March 21—SAFE Credit Union Performing Arts Center—Sacramento, CA
March 22—Dolby Theatre—Los Angeles, CA
March 24—Encore Theater at Wynn Las Vegas—Las Vegas, NV
March 25—Encore Theater at Wynn Las Vegas —Las Vegas, NV
March 27—Ellie Caulkins Opera House—Denver, CO
April 3—The Midland Theatre—Kansas City, MO
April 4—The Midland Theatre—Kansas City, MO
April 6—State Theatre—Minneapolis, MN
April 7—State Theatre—Minneapolis, MN
April 8—State Theatre—Minneapolis, MN
April 10—The Riverside Theater—Milwaukee, WI
April 11—The Riverside Theater—Milwaukee, WI
April 13—Chicago Theatre—Chicago, IL
April 14—Chicago Theatre—Chicago, IL
April 17—The Kentucky Center—Louisville, KY
April 18—The Kentucky Center—Louisville, KY
April 19—Peoria Civic Center Theater—Peoria, IL
April 21—Stifel Theatre—St. Louis, MO
April 22—Stifel Theatre—St. Louis, MO
April 24—Orpheum Theatre—Memphis, TN
April 25—Orpheum Theatre—Memphis, TN
May 5—Old National Events Plaza—Evansville, IN
May 6—Old National Events Plaza—Evansville, IN
May 8—Ryman Auditorium—Nashville, TN
May 9—Ryman Auditorium—Nashville, TN
May 10—Ryman Auditorium—Nashville, TN
May 12—Aronoff Center—Cincinnati, OH
May 13—Aronoff Center—Cincinnati, OH
May 15—Embassy Theatre—Ft. Wayne, IN
May 16—Embassy Theatre—Ft. Wayne, IN
May 17—Palace Theatre—Columbus, OH
May 19—Peace Center—Greenville, SC
May 20—DPAC —Durham, NC
May 22—Benedum Center—Pittsburgh, PA
May 23—Benedum Center—Pittsburgh, PA
May 25—Playhouse Square—Cleveland, OH
May 26—Playhouse Square—Cleveland, OH
June 2—The Lyric—Baltimore, MD
June 3—Kimmel Cultural Campus—Philadelphia, PA
June 5—Beacon Theatre—New York, NY
June 6—Beacon Theatre—New York, NY
June 7—Beacon Theatre—New York, NY
June 10—Boch Center – Wang Theatre—Boston, MA
June 11—Providence Performing Arts Center—Providence, RI
June 13—Palace Theatre—Albany, NY
June 14—Landmark Theatre—Syracuse, NY
June 16—Shea’s Performing Arts Center—Buffalo, NY
June 17—Masonic Temple Theatre—Detroit, MI
June 19—Massey Hall—Toronto, ON
June 21—DeVos Performance Hall—Grand Rapids, MI
June 23—Morris Performing Arts Center—South Bend, IN
June 24—Morris Performing Arts Center—South Bend, IN

Be A Part Of The Grateful Web

Contact US
Progressive jam giants Umphrey’s McGee‘s return to Las Vegas for the seventh installment of the massively popular UMBowl production was marked once again by a stand-out tour closing dual evening extravaganza where all stops were pulled out and the power given directly to the fans, for better or for worse.
On June 24, Round Records & ATO Records will release GarciaLive Volume Six: July 5, 1973 – Jerry Garcia & Merl Saunders, the latest installment of the celebrated GarciaLive archival series. The three-CD set was recorded at the 200 capacity Lion’s Share club formerly located in the small town of San Anselmo, CA, just 20 miles north of San Francisco. The performance features Jerry Garcia performing with friend, mentor and legendary keyboardist/vocalist Merl Saunders. The duo is joined by drummer Bill Vitt and bassist John Kahn, who soon became a lifelong Garcia collaborator.
Progressive jam giants Umphrey’s McGee‘s return to Las Vegas for the seventh installment of the massively popular UMBowl production was marked once again by a stand-out tour closing dual evening extravaganza where all stops were pulled out and the power given directly to the fans, for better or for worse.
COPYRIGHT © 1995 – 2022 GRATEFUL WEB, INC. ALL RIGHTS RESERVED.

source

Make sure you find all the Secret Cache locations in Gotham Knights to uncover Bruce Wayne’s audio recordings.
Gotham Knights is an open-world RPG game set in Gotham City that follows the story of the Batman Family. Batman is dead, and it is up to Red Hood, Robin, Batgirl, and Nightwing to join forces and protect Gotham from criminals who are slowly starting to realize that Bruce Wayne is gone. There are many collectibles to be hunted for within the game.
RELATED: Gotham Knights: Case 2.2 Powers Club Shadow Bird Puzzle Solution
Bruce Wayne hid audio recordings around Gotham City before his death. There are 12 recordings to be collected, and you will unlock the Gotham City Confidential trophy/achievement upon doing so. Although the trophy list refers to these as Bruce’s recordings, it is important to note that they are referred to as Secret Caches in the game.
Upon reaching the locations of each of the Secret Caches, you will have to complete a time trial to unlock the audio recording. There will be a floor plate your character will have to step on while you hold the button shown on the screen to start the trial. A timer will start, and you will have to reach all the goals before it runs out.
The path will be marked with blue lights to indicate where you need to go to complete the challenge. This will usually just involve Parkour and quick reaction times, but can prove difficult to complete if you are impatient or rush the trial. Upon completing the trial, a marker will appear to indicate the location of the Cache for you to collect.
The first Secret Cache can be found in the Financial District after completing Main Case 01. Upon collecting this cache, numbers two to five will spawn on the map once you have returned to Belfry.
The second Secret Cache can also be found in the Financial District, but this is only after you have collected the first.
The third Secret Cache can be found in The Cauldron.
The fourth Secret Cache can be found in the Southside.
The fifth Secret Cache can be found within Old Gotham. Upon collecting the fifth Cache, you will need to return to Belfry. Secret Cache numbers six to nine will then spawn on the map for you to access.
The sixth Secret Cache will be on Tricorner Island.
The seventh Secret Cache can be found in the Bowery.
The eighth Secret Cache will be in Otisburg.
The ninth Secret Cache can be found in West End. After collecting this one, return to Belfry to allow the final three Secret Cache locations to spawn on the map.
The tenth Secret Cache will be in Gotham Heights.
The eleventh Secret Cache can be found in Bristol.
And the twelfth and final Secret Cache can also be found in Bristol.
If you are looking to find all the locations for each Secret Cache in the shortest amount of time possible, then it may help to find the special Secret Cache location. You will need to go to this one first, and it can be located between the Financial District and The Cauldron (2.1 – AKA Oswald Cobblepot).
Your character will have to step on a golden plate trigger and once you have done so, you will unlock all the locations on the map. Although, you will still have to return to Belfry after completing each batch. Return to Belfry after stepping on this gold plate and completing the time trial, and the first four location icons will spawn on the map.
NEXT: Games To Play If You Like Gotham Knights
A female gamer who enjoys FPS, Puzzle and Indie games. She has a Bachelor’s degree majoring in English Studies and Ancient Cultures and is currently completing her Honors Degree in Ancient Studies. She has a passion for writing and takes up freelancing as a means to satisfy the urge to write. Likewise, she has written for online publishers such as Scholaroo and The South African.

source

Sign up to read our regular email newsletters
In 1992, a TV show petrified millions of UK viewers with a “live” investigation of a paranormal event. What happened next? Its writer, Stephen Volk, explains
By George Bass

Ghostwatch
via Stephen Volk
Ghostwatch
Writer: Stephen Volk
Director: Lesley Manning
Blu-ray release 31 October
It was the night of Halloween, 1992. Millions of British TV viewers had tuned into a BBC TV programme called Ghostwatch to watch a live investigation into a haunted house. Hosted by Michael Parkinson and Sarah Greene, the show featured parapsychologists and electronics engineers who were certain their recording equipment was detecting something sinister. Viewers were told that they had taken part in a “national séance” that had unleashed a great evil via the TV signal.
This 1992 …
Continue reading
Subscribe today
with a 12-week introductory offer
No commitment, cancel anytime*
Offer ends 19/10/2022. *Cancel anytime within 14 days of payment to receive a refund on unserved issues.
Existing subscribers, please log in with your email address to link your account access.
Inclusive of applicable taxes (VAT)

source

Photo credit: CPAC; Gorgo (Wikimedia)
Phone calls suggest conservative ‘political fixer’ may have acted as a ‘foreign intelligence source’
The Royal Canadian Mounted Police appear to have had knowledge over two years ago about an alleged plot to entrap former Calgary Mayor Naheed Nenshi, a case Calgary Police announced this week is the focus of an active criminal investigation.
The alleged entrapment plot, called “Operation Peacock,” involved several conservative insiders, developers and a self-described “political fixer” named David Wallace.
EXCLUSIVE: Calgary conservatives plotted to entrap and oust mayor Nenshi in “big store con” scheme https://t.co/r6HiC8oHSd
— CANADALAND (@CANADALAND) October 31, 2022

According to an investigation by Canadaland, the goal of the alleged plot was to frame Nenshi and members of Calgary city council by capturing compromising material showing them soliciting sanctioned Russian money:
“Wallace said that his plan was what’s known as a ‘big store con’, in which he presented himself to the mayor and city councillors as a representative of billions of dollars in Russian investment money. Eventually, he intended to bug rented office space with recording devices, and invite the mayor to a meeting there. He also hoped to convince the mayor and city councillors to sign a letter to the Russian Embassy in Ottawa requesting a bilateral trade agreement. Ultimately, he hoped to capture Mayor Nenshi soliciting money that would be revealed as dark funds from the list of sanctioned Russian oligarchs. He would then leak the tape to the press.”
Russian officials were invited to the city in 2019 in a letter to the Russian Embassy in Ottawa signed by two Calgary city councillors, Sean Chu and Joe Magliocca – but the visit never happened.
On Thursday, Calgary Police launched a criminal investigation into the alleged plot targeting Nenshi.
Former mayor Naheed Nenshi has issued this statement re: Canadaland story. #yyccc pic.twitter.com/g7jGfJO9QD
— Scott Dippel (@CBCScott) October 31, 2022

Except Calgary Police are not the first law enforcement agency to be aware that Nenshi was the target of an alleged plot.
According to audio recordings obtained by PressProgress, an RCMP investigator had been tipped off about clandestine efforts to “get rid of Nenshi” as early as 2020.
Audio of a purported phone conversation between Wallace and RCMP Cpl. Steve Morrissey suggests the self-described “political fixer” had also been acting as an intelligence source for the RCMP. Much of the phone call, which was recorded by Wallace, focuses on overseas business activities of a number of Wallace’s contacts before turning to domestic topics.
“What happened with this Calgary stuff with Nenshi?” Morrissey asks Wallace in the recording.
“Nothing’s happened with it,” Wallace replies. “I walked away. I walked away.”
“I’ve had Prem, again, I’ve asked her not to contact me, I don’t answer the phone, I haven’t spoken to her probably two months now, she keeps sending me messages: ‘Bill Turnbull, this person, they want to spend the money now. They found the money to pay you off. Now they want to get rid of Nenshi’.”
“I can actually send you a copy of some of the messages Prem sent me if you’d like. You can read them yourself.”
Morrissey replied: “You can show me next time we get together.”
Local conservative activist Prem Singh and Calgary businessman Bill Turnbull both confirmed to PressProgress they were aware of “Operation Peacock” at the time, though the exact details of their involvement remain to be confirmed.

Contacted at the phone number associated with the recording, Cpl. Morrissey confirmed his identity but declined comment, stating he was “off-duty” and hung up.
RCMP K-Division confirmed Morrissey still works for them but said that as a standard policy, it cannot not comment on information that could be subject to investigations.
“Only in the event that an investigation results in the laying of criminal charges would the RCMP confirm its investigation, the nature of any charges laid and the identity of the individual or individuals involved,” K-Division spokesperson Fraser Logan told PressProgress.
“However, I can confirm that Cpl. Steve Morrissey is a member of K-Division RCMP.”
Wallace claims he has occasionally acted as a “source of foreign intelligence” to the RCMP.
Wallace says he was first introduced to the RCMP in 2020 by local conservative political activist Prem Singh and had “regular contact” with them until early 2021.
“I introduced David to the RCMP that I had spoken to,” Prem Singh told PressProgress. “Then they wanted to talk to him on their own, and apparently told him not to talk to me. And then the RCMP guy basically stopped talking to me.”
Singh said she was not certain about why the RCMP wanted to talk to Wallace, but said she assumed it was in relation to issues that arose during an RCMP investigation looking into the 2017 UCP leadership race.
That meeting is detailed in an audio recording of another purported phone call that took place on August 4, 2020.
“I’m Corporal Steve Morrissey and I work in a unit in Alberta,” Morrissey says in the audio recording. “The RCMP identifies themselves as divisions, so I work in K-division.”
Both Singh and Wallace confirmed to PressProgress that they recalled participating in this call. After making introductions, Singh hopped off the call, leaving Wallace and Morrissey to have a discussion in private.
“So, I work intelligence,” Morrissey tells Wallace.

“How it goes with the off-the-record conversations is I have to kind of assess the information you’re going to give me, because I’m not offering any confidentiality at this point, any privilege or nothing, because I can’t, I need to know the veracity of what you’re telling me,” Morrissey says.
“Criminal Code bounds me to protect identities on confidential informants, but I can’t make you a confidential informant without having a secondary member, we test the veracity of the information you’re giving me,” Morrissey explains. “And then I’ll look at extending your privilege.”
“So it’s kind of like, give me a sneak peek of what we’re going to talk about, I go back, and then the way it kind of works is, in order for me to protect your identity.”
In the same recording, Morrissey discloses he previously worked on the “UCP investigation file.”
“I was transferred to another unit in October. But I’m still attached to this file because I have a lot of due interest in the file.”
First reported by PressProgress in 2019, the RCMP launched investigations looking into allegations of identity theft during the 2017 UCP leadership race as well as a so-called “kamikaze” campaign run by allies of former Premier Jason Kenney.
Earlier this year, the Toronto Star, who relied on the same source documents from Wallace, reported that RCMP investigators had potentially compromised the investigation after leaving voice recorders at the home of Allan Hallman, an associate of Jason Kenney.
According to the Star, the RCMP left the voice recorder at Hallman’s home following an interview in connection to the investigation on January 21, 2020.
Allan Hallman’s son, Chad, confirmed to Canadaland he had initial discussions with Wallace about Nenshi with Wallace and connected him with Singh. Chad Hallman, who worked as one of Kenney’s UCP political staffers, denied any involvement in raising money for the alleged plot.
Earlier this year, PressProgress reported Wallace was also involved in an effort to flow Russian money into Ontario real estate developments in 2018 amid US sanctions.
Frank Klees, a trusted adviser to Ontario Premier Doug Ford, confirmed relaying a backchannel message to Ford’s former Chief of Staff Dean French about setting up meetings between Russian officials and Ford’s government.
 
 
Update: This story has been updated to include a clip from the August 2020 audio recording between Wallace and Morrissey.
 
Our journalism is powered by readers like you.

We’re an award-winning non-profit news organization that covers topics like social and economic inequality, big business and labour, and right-wing extremism.
Help us build so we can bring to light stories that don’t get the attention they deserve from Canada’s big corporate media outlets.
 
We strive to avoid typos and inaccuracies. However, on occasion we make mistakes. We value your contributions and help in correcting them.

We value accuracy. If you spot an error, use this form to notify our editorial team. Please see our corrections policy for more info.
Your Name

Your Email

City, Province and Country


Title of Article


Describe the typo or inaccuracy:


2 days ago
5 days ago
6 days ago
3 weeks ago
10 months ago
1 year ago
1 year ago
Why do newspapers always seem to have a business section but never a labour section? We‘ve launched a FREE weekly email newsletter covering labour news in Canada.
PressProgress is a news division of the Broadbent Institute
PressProgress journalists are proudly unionized with UFCW Canada Local 1006-A

PressProgress Newsroom |
Station B, PO Box 1273 Ottawa, Ontario K1P 5R3

source

Thanks for contacting us. We've received your submission.
By Riley Cardoza
Submit
Δ
November 1, 2022 | 2:31pm
Master P’s daughter Tytyana Miller’s cause of death has been revealed.
The 25-year-old died of accidental fentanyl intoxication at a private residence, according to the Los Angeles County Department of Medical Examiner-Coroner (via People).
The rapper, 52, mourned Miller’s death in a May 29 Instagram post, two days after she died.
“Our family is dealing with an overwhelming grief for the loss of my daughter Tytyana,” he wrote at the time.
“Mental illness & substance abuse is a real issue that we can’t be afraid to talk about. With God, we will get through this. #MyAngel.”
Master P’s son Romeo Miller also mourned Tytyana via Instagram at the time.
“Although this is sad times, I’m forever grateful for the memories I did have with my amazing sister,” the actor told his followers. “Love on your loved ones, life is short.”
Two months later, Master P spoke of his “hard” grieving process during a “CBS This Morning” appearance.
Getty Images via Getty Images
WireImage
Getty Images
WireImage
View Slideshow
“I feel like, going to my daughter’s funeral, I feel like I went to my own funeral,” he said in July. “I don’t even want to go to no more funerals.”
Master P noted that Tytyana had “a lot of life left in her,” adding, “She was a happy woman that knew that tomorrow she could do something better. … I love her. I think about her every day.”
Prior to her death, Tytyana and her family members were vocal about her addiction struggles on “Growing Up Hip Hop.”
“When people hide that they have a problem, they’ll never overcome it,” Master P said in a 2016 episode. “She was like, ‘Dad, I want to be better. I want to get myself together.’”
Read Next
Leni Klum, 18, comments on controversial lingerie shoot wi…

This story has been shared 203,935 times. 203,935
This story has been shared 79,275 times. 79,275
This story has been shared 58,472 times. 58,472
This story has been shared 39,360 times. 39,360
Follow us
© 2022 NYP Holdings, Inc. All Rights Reserved Terms of Use Privacy Notice Your Ad Choices Sitemap Your California Privacy Rights Do Not Sell My Personal Information
Your California Privacy Rights Do Not Sell My Personal Information

source

WATCH LIVE
Some New York City Uber drivers will be allowed to record rides on their phones as part of a new pilot program.
NEW YORK — Uber drivers will take fewer left turns and soon be able to record rides through their smartphones as the ride-share company makes changes geared toward safety in the driver's seat, officials said.
The company announced Thursday that it is launching a new pilot program in which certain drivers in three U.S. cities and in Brazil can use their phone's front-facing camera to record audio and video during trips.
"Recording can just improve and make interactions on the Uber platform a little bit more comfortable because everyone knows that they're going to be held accountable for their actions," Rebecca Payne, group product manager on safety at Uber, said in an interview with ABC News.
The pilot is an expansion of the company's already existing audio recording feature, in which drivers and passengers can both opt to record audio of trips.
Passengers will be notified after requesting a trip that their driver will be video-recording the ride, Uber said. If they don't feel comfortable being recorded, passengers can cancel at no extra cost.
"We've seen many instances where this technology has helped us determine the best course of action after a safety incident, and the majority of riders and drivers in the pilot cities told us this feature helped them feel safer when using Uber," the company said in a news release.
The new video recording technology will be available to select drivers in Cincinnati, Louisville and New York City as well as Santos and Joo Pessoa in Brazil, Uber said.
When asked about privacy concerns, Uber said the driver's video recordings will be encrypted and stored directly on the driver's device but inaccessible even to them.
"No one can access it … even Uber can't access it," Payne said. "If nothing bad happens on the trip, that recording will essentially just disappear after seven days."
But should a safety incident occur during a trip, the driver could attach the encrypted video file to the safety report sent to Uber. Once the company had the report, the file would then be decrypted and a trained safety agent would review it to help determine what occurred, the Uber said.
The company also announced Thursday it will update its in-app navigation software to suggest drivers make fewer left turns. According to the National Highway Traffic Safety Administration (NHTSA), 22% of crashes involved a car making a left turn at an intersection.
"Essentially what it does is when a rider puts in the destination, our algorithm and our navigation will choose a routing to reduce those lectures as much as possible without adding any additional time or cost for the trip," Kristin Smith, Uber's head of road safety policy, told ABC News. "We're hopeful that this will be one of those tech interventions that can help to really improve road safety."
ABC News' Sam Sweeney contributed to this report.
ALSO READ | Thousands of officers to patrol subway system as focus shifts to mentally ill

———-
* Get Eyewitness News Delivered
* More New York City news
* Send us a news tip
* Download the abc7NY app for breaking news alerts
* Follow us on YouTube
Submit a News Tip
NY governor candidates make final push to sway undecided voters
Watch Sunday: Election 2022 preview special
38 injured in high-rise fire sparked by a lithium-ion battery: FDNY

source

Thank you for visiting nature.com. You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.
Advertisement
Nature Communications volume 13, Article number: 6529 (2022) Cite this article
3151 Accesses
20 Altmetric
Metrics details
Age is the leading risk factor for prevalent diseases and death. However, the relation between age-related physiological changes and lifespan is poorly understood. We combined analytical and machine learning tools to describe the aging process in large sets of longitudinal measurements. Assuming that aging results from a dynamic instability of the organism state, we designed a deep artificial neural network, including auto-encoder and auto-regression (AR) components. The AR model tied the dynamics of physiological state with the stochastic evolution of a single variable, the “dynamic frailty indicator” (dFI). In a subset of blood tests from the Mouse Phenome Database, dFI increased exponentially and predicted the remaining lifespan. The observation of the limiting dFI was consistent with the late-life mortality deceleration. dFI changed along with hallmarks of aging, including frailty index, molecular markers of inflammation, senescent cell accumulation, and responded to life-shortening (high-fat diet) and life-extending (rapamycin) treatments.
Aging manifests itself on multiple levels of the organism organization. Accordingly, diverse sets of physiological state variables, such as DNA methylation patterns or blood composition markers^1,2,3,4,5, can be used to quantify aging. State-of-the-art biomarkers of aging came out from supervised models that require chronological age or mortality age as labels for training (see, e.g., the most recent⁶). Alternatively, the number of health deficits accumulated⁷ or composite frailty indices^8,9 correlate with the disease burden and remaining lifespan of humans and animals^9,10 and find applications in laboratory experiments and clinical trials of anti-aging interventions^3,11,12. Future use of aging biomarkers will depend on a better understanding of the connection between aging hallmarks, biological age predictors, frailty indices, and the remaining lifespan.
Here, we demonstrate a combination of modern machine learning techniques and approaches borrowed from the dynamic systems theory working together to identify quantitative rules governing manifestations of aging and their relation to all-cause mortality in big longitudinal biomedical data. We start by presuming that aging is a particular case of the dynamics of a complex system unfolding near a bifurcation or a tipping point on the boundary of a dynamic stability region. Under the circumstances, the organism state fluctuations should be driven by the dynamics of a very few, if not a single collective feature having the meaning of the order parameter corresponding to the unstable phase^13,14,15. Examples of such dimensionality reduction in biological sciences include development¹⁶, aging, and mortality acceleration^5,17,18.
We built upon this theoretical concept and produced a deep artificial neural network composed of a denoising autoencoder (AE) and an auto-regressive (AR) model — a computational metaphor for the stochastic dynamics of the order parameter. To test the approach, we used the network to produce a descriptor of aging based on complete blood count (CBC) measurements from the largest open-access source of phenotypic data, the Mouse Phenome Database (MPD)¹⁹. The network output variable, hereinafter referred to as the dynamic frailty indicator (dFI), is the best numerical approximation for the order parameter associated with the organism state disintegration and hence aging from any given data. We demonstrated that the dynamics of dFI drives the mortality acceleration and explain late-life deviations from Gompertz mortality law. The dFI exhibited the most desirable properties of a biological age marker: it increased exponentially with age, predicted the remaining lifespan of the animals, correlated with multiple hallmarks of aging, and responded to interventions known to accelerate (high-fat diet) and decelerate (rapamycin) aging in mice.
We used the largest publicly available longitudinal phenotypic data source, the MPD¹⁹. To achieve the best possible compatibility with earlier studies, we scanned the database records to maximize the number of available measurements common to those used in constructing the physiological frailty index (PFI) in ref. 3. As a result, we chose a subset of 12 CBC measurements from nine datasets, altogether including 6693 animals (see Supplementary Data 1 for the full list of datasets used for training the models in this work).
To visualize the 12-dimensional CBC data from the MPD, we performed principal component analysis (PCA). PCA of the MPD slice representing fully grown animals (exceeding the age of 25-week-old, see Methods for the details regarding the relevant age-range determination) turned out to be particularly simple. In this case, most of the variance in the data (31%) is explained by the first PC score, z₀, with the subsequent PC scores (z₁, z₂, etc., each explaining 19%, 16%, etc. of the total variance in the data, respectively). The hierarchical clustering of the CBC features suggested that the first two PC scores could be predominantly connected with the red and white blood cell counts, as shown in Fig. 1a, respectively.
a Clustering of CBC features and PC scores in the training dataset. The colors represent Pearson’s correlation coefficient (absolute value) as indicated by the scale on the right. b The graphs represent the average of the PC scores in subsequent age groups (the error bars are standard deviations). The inset shows that the variance for all PC scores increases with age. Two-sided p values were calculated for the Pearson correlation coefficient in the sample size of n = 1448 animals.
Only the first PC score, z₀, demonstrated an appreciable correlation with the chronological age (the corresponding Pearson’s correlation coefficients were r = 0.59 (p < 10⁻⁵) and r = 0.08 (p = 0.003) for the first two PC scores, respectively (Fig. 1b). The variance of the PC scores increased with age (see the inset in Fig. 1b), which is a signature of a stochastic process.
The association of the fluctuations along the dominant PC with the slowest dynamic process (aging, in our case) has deep roots in the dynamic systems theory and critical phenomena¹⁴. The transition from stability to instability in networks with the network graphs not possessing specific symmetries is typically associated with co-dimension 1 (or saddle-node) bifurcations^20,21. Such transitions are characterized by the loss of stability along a single direction in the state vector space, approximately coinciding with the first principal component. In contrast, contributions from all other principal components remain stable (and hence age-independent). Accordingly, over sufficiently long timescales, the fluctuations of physiological indices (such as CBC features), x_i, are expected to follow the dynamics of the order parameter, z, associated with the instability: x_i ≈ b_iz + ξ_i. Here ξ_i is noise, b_i is a vector, and the integer index i enumerates the measured features.
Close to the tipping point, the dynamics of the physiological state is slow and approximately linear. Hence the variable z satisfies the stochastic Langevin equation with the higher-order time derivative terms neglected:
Here the linear term, αz, on the right side of the equation represents the effect of the regulatory network stiffness governing the organism’s responses to small stresses and producing slight deviations of the organism state from its most stable position. The following term, gz², represents the lowest-order non-linear coupling effects of regulatory interactions.
The stochastic forces f represent external stresses and the effects of endogenous factors not described by the effective Eq. (1). Naturally, we assume that random perturbations of the organism state are serially uncorrelated, so that (langle f(t)f({t}^{prime})rangle sim Bdelta ({t}^{prime}-t)), where B is the power of the noise, δ is the Dirac’s delta-function, and 〈…〉 stands for averaging along the aging trajectory. Eq. (1) is a mathematical relationship between the rate of change of the organism state variable, (dot{z}=dz/dt), on the left side of the equation, and the effects of deterministic (αz, gz²) and stochastic forces (f), on the right side.
Depending on the sign of the stiffness coefficient, α, the organism state may be dynamically stable (if α < 0) or unstable (if α > 0). In the latter case, minor deviations of the organism state get amplified exponentially over time, so no equilibrium is possible. In this case, the solution of Eq. (1) describes an aging organism. Typically, α is small, and hence, the evolution of the physiological indices exhibits hallmarks of critically: the dynamics of the order parameter are slow (critical slowing down), whereas the fluctuations of the physiological state following the variations in z are large, ({z}^{2} sim Bexp (2alpha t)/2alpha gg 1) (critical fluctuations).
Very early in life, the deviations from the critical point are small, and the evolution of the organism state is dominated by diffusion. Later in life, the linear term takes over such that the deviations from the youthful state accelerate exponentially:
where (bar{z} sim {(B/alpha )}^{1/2}) and z₀ are constants representing the accumulated early effects of random and deterministic forces, respectively.
Finally, once the order parameter is sufficiently large, z ≳ Z = α/g, the non-linear terms take over, the disintegration of the organism state proceeds at a rate greater than exponential, and the animal dies in a finite time. Eq. (1) can be solved to obtain the simple analytical expression for the fraction of animals surviving up until the age t: (S(t), approx ,{{{{{{{rm{Erf}}}}}}}}(Z/bar{z}exp (-alpha t))), where Erf(x) is the error function¹⁸. The theoretical survival function is different from that of the Gompertz mortality law implying the exponential acceleration of mortality at all ages.
The all-cause mortality in the model, (M(t)=-dot{S}(t)/S(t)), increases quickly at first in the diffusion-dominated regime and then slows down at the age, approximately coinciding with the average lifespan (bar{t}=1/alpha log (Z/bar{z})). In the eldest animals, the disagreement with the Gompertz mortality law is the strongest: the mortality in the model decelerates and reaches the plateau at (M(t, gg ,bar{t}), approx ,alpha).
Finally, we observe, that the sample co-variance matrix of the organism state variables in the unstable regime is dominated by the dynamics of the order parameter:
where 〈…〉 stands for sample average. In the weak nonlinearity limit, (alpha bar{t}, gg ,1), PCA of such signal would reveal the loading vector corresponding to the first PC score coinciding with b_i. The first PC score z₀ = (b, x) is then the approximation to the order parameter z ≈ z₀ itself.
Aging at criticality may be a useful theoretical concept. However, it does not provide a practical algorithm for identifying the order parameter other than an approximate application of PCA. To characterize the aging process from the measurements, we analyzed the longitudinal aging trajectories of CBC measurements from 6,693 samples in nine MPD datasets (see Methods and Supplementary Data 1) with the help of an artificial neural network. The algorithm included a combination of a deep auto-encoder (AE) for the dimensionality reduction and a simple AR model for the fit of the network representations of the subsequent measurements onto solutions of Eq. (1) (AE–AR; see Methods).
The AR problem’s solution, z, is the output of the algorithm and is the best fit estimation of the order parameter associated with the disintegration of the organism state and hence aging from the available data. Below we establish the associations of z with hallmarks of aging, frailty, and mortality. Accordingly, we chose to refer to z as the dFI.
The performance of the AR model was demonstrated by plotting the auto-correlation property of dFI, which is the correlation between dFI values measured along aging trajectories of the same mice at age points separated by 14 and 28 weeks in the test dataset MA0072 (Fig. 2). The corresponding Pearson correlation coefficients between the respective age-adjusted dFI estimations were r = 0.71 (p < 0.001) and r = 0.70 (p < 0.001). The dFI auto-correlations were better than the auto-correlations of the first PC score z₀ for the same mice, see Supplementary Fig. 2; the corresponding Pearson’s correlation values were r = 0.55 (p < 0.001) and r = 0.71 (p < 0.001) for 14- and 28-week time lags, correspondingly.
Correlation between age-adjusted dFI values across sampling intervals Δt of 14 (blue circles, animals n = 40) and 28 (orange squares, animals n = 19) weeks in the validation MA0072 dataset. Two-sided p values were calculated for the Pearson correlation coefficient.
A semi-quantitative hierarchical clustering of the CBC features’ co-variances in the test dataset produced correlations across the features associated with the immune system (white blood cell counts and the related quantities), metabolic rate/oxygen consumption (red blood cell counts and hemoglobin concentrations), and an apparently independent subsystem formed by platelets (Supplementary Fig. 6).
dFI increased with the age of the animals both in the test and in the training sets (Fig. 3 and Supplementary Fig. 3, respectively). As expected from the qualitative solution of Eq. (1), dFI increased up to the age corresponding to the average animal lifespan (~100 weeks in our case). We performed an exponential fit in the form of Eq. (2) on the data from the test datasets (excluding animals that lived longer than the strain’s average lifespan and animals at the end of their life from the dataset MA0073). The calculation returned dFI growth exponent of α = 0.02 per week.
The dependence of the dFI from age in the validation datasets from the experiments: MA0071 (males, orange diamonds), MA0071 (females, blue circles), and MA0072 (green triangles). The black dashed line is the exponential fit in the age groups younger than the average lifespan of NIH Swiss mice (indicated by the vertical grey dashed line). Red stars mark the average dFI in age-matched groups of frail animals from the MA0073 cohort. All data are presented as Mean ± SEM.
The saturation of the dFI beyond the average lifespan in the training and test datasets revealed a limiting value that is apparently incompatible with the animals’ survival. To highlight this possibility, we plotted the dFI ranges from a separate cohort of “unhealthy” mice from the MA0073 experiment, representing the animals scheduled for euthanasia under lab requirements (Fig. 3, red stars).
The character of the dFI acceleration with age and the dFI doubling rate matching the mortality acceleration rate (equal to 0.037 per week²²) are good indicators of the association between dFI and mortality. The observation could be further supported by computing the Spearman’s rank correlation between the dFI “acceleration” (i.e., the difference between the dFI of an animal and its mean value in the corresponding age- and sex-matched cohorts) and the order of the death events among the animals of same age and sex (see Table 1). We obtained significant correlations between the dFI and the remaining lifespan for all cohorts.
As a benchmark, we also produced a supervised Cox proportional hazards (PH) model predicting animal survival in the same data. Notably, the age- and sex-adjusted dFI predicted the remaining lifespan equally well or marginally better than the log-hazard ratio predictor from the supervised model trained in the same data (c.f. the rows corresponding to “dFI” and “HR_CBC” in Table 1).
The dFI predicted remaining lifespan later in life better than body weight (BW) or insulin-like growth factor 1 (IGF1) serum level. Both factors were previously shown to be associated with mortality in refs. 23 and 24. As pointed out in ref. 24 and checked here, the concentration of IGF1 in serum was significantly associated with lifespan (r = − 0.28, p = 0.008) only in one cohort of younger, 26-week-old male mice. According to ref. 23 and our calculations, BW is better associated with mortality in the youngest animals (age 26 and 52 weeks).
The number of animals with recorded lifespans in MPD was insufficient to understand the deviations of mortality from Gompertz mortality law, as predicted by the aging at criticality model at the most advanced ages. Fortunately, a quantitative analysis of late-life mortality could be performed using the data from the non-treated controls in a large experiment involving thousands of mice²⁵. Gompertz mortality fits produced the following estimations for the lifespan, (bar{t}), the Gompertz exponent, α and the initial mortality rate, M₀ separately for males: (bar{t}=100.3) weeks, α = 0.0385 ± 0.0002 per week, M₀ = (4.1 ± 0.1) ⋅ 10⁻⁴ per week; and for females: (bar{t}=115.8) weeks, α = 0.0568 ± 0.0003 per week, M₀ = (3.4 ± 0.1) ⋅ 10⁻⁵ per week.
Late in life in both sexes, the empirical survival curves fall slower than the predictions from the Gompertz mortality law (see the solid blue and the orange dashed lines in Fig. 4a and Supplementary Fig. 4a, respectively). This is a signature of the deceleration of mortality. The blue lines in Fig. 4b and Supplementary Fig. 4b are the empirical mortality curves. The grey dashed lines correspond to the theoretical expectation for the level of mortality at the plateau, corresponding to the theoretical limiting mortality (M(t, gg ,bar{t})=alpha), where α is the mortality rate doubling rate from the Gompertz fit.
a The Kaplan-Meier survival curve (solid blue line) in female mice cohorts from ref. 25. The orange dashed line represent the best Gompertz fits. b The Nelson–Aalen estimator of total mortality (solid blue line) and the 95% confidence intervals are filled in blue. The grey dashed line corresponds to the mortality level according to the theoretical prediction (M(t, gg ,bar{t})=alpha).
The leveling off of mortality is better pronounced in the female cohorts (Fig. 4) and is consistent with the theoretical prediction of the late-life plateau mortality in our model. In the male cohorts, the animals expire before the plateau is reached (Supplementary Fig. 4), possibly due to the elevated background mortality earlier in life between weeks 40 and 100. Nevertheless, the maximum observed mortality is still not very far from (and is definitely of the same order as) the predicted limiting value even in male cohorts.
To further validate dFI as a biomarker of aging, we examined its relation to the PFI, a quantitative measure of aging and frailty recently proposed in ref. 3. dFI and PFI were found to be strongly correlated (Pearson’s r = 0.64, p < 0.001), Fig. 5a. PFI is a composite frailty score and also depends on CBC measures for its determination. However, the calculation of PFI involves more factors, including traditional measures of frailty in animals and humans, such as grip strength, cardiovascular health, inflammation markers, etc. Remarkably, the correlation between PFI and dFI remained significant after adjustment for sex and age (Supplementary Fig. 5, Pearson’s r = 0.59, p < 0.001).
a Correlation between the dFI and the physiological frailty index (PFI). Colors represent animals from the test dataset, where blue, orange and green circles are females in MA0071, males in MA0071 and males in MA0072, respectively. b Volcano plot representation of the dFI correlation with the extended set of biomarkers in the test datasets MA0071 and MA0072. Features with correlation above and below significance level p < 0.001 are shown with grey and blue circles, respectively. The most significant correlations (excluded dFI components) were between dFI and C-reactive protein (CRP), red cell distribution width (RDW), body weight (BW), and murine chemokine CXCL1 (KC). Two-sided p values were calculated for the Pearson correlation coefficient and the sample size of n = 282 animals.
As illustrated in Fig. 5b and Supplementary Fig. 6, 7, we observed that the dFI was significantly associated with an extended set of CBC features across independent functional subsystems (most notably, but not limited to, myeloid cell lineage). The correlation between dFI and myeloid cell features was less profound in the training set, involving multiple strains (Supplementary Fig. 7). The correlation coefficient is a measure of the response of dFI to individual CBC features variation and is different (sometimes even of opposite sign) in various mouse strains (Supplementary Fig. 8). The variation of the associations of individual features and dFI would be a significant challenge to any linear model and is a demonstration of the non-linear character of the AE.
The dFI was strongly associated with red blood cell distribution width (RDW) and BW (Fig. 5b), known predictors of frailty in both mice²⁶ and humans²⁷. dFI was also strongly associated with levels of C-reactive protein (CRP, r = 0.39, p < 0.001) and the murine chemokine CXCL1 (KC, r = 0.28, p < 0.001), both of which are known markers of systemic inflammation and mortality^28,29.
Aging is associated with an increasing burden of senescent cells^30,31,32, widely considered a source of chronic sterile systemic inflammation, “inflammaging”³³. Senescent cells (SC) are commonly detected in vivo as a population of p16/Ink4a-positive cells accumulated with age recognized by the activity of p16/Ink4a promoter-driven reporters³⁴. We utilized homozygous p16/Ink4a reporter mice with one p16/Ink4a allele knocked in with firefly luciferase cDNA³⁵. Figure 6a shows the correlation between animal age and the presence of senescent cells, as measured by the flux from p16/Ink4a promoter-driven luciferase activity (r = 0.54, p = 0.008).
Total flux (TF) in log scale represents p16-dependent luciferase reporter activity, is a quantitative indicator of senescent cells burden, and shows statistically significant correlations with age (Pearson’s r = 0.54, two-sided p = 0.008, n = 23 animals) (a) and with dFI (Pearson’s r = 0.69, two-sided p = 0.0003, n = 23 animals) (b) in old mice (>50 weeks). The colorbar in b represents animal age in weeks. dFI responds to the lifespan-modifying effect of high-fat diet (HFD): the dFI values (dots) were obtained late in life (at week 78) for male (c) and female (d) mice fed with RD or HFD. The horizontal bar and the dashed lines indicate the mean values in the groups and for all animals, respectively. dFI was significantly higher in males with HFD (n = 7 animals) vs RD (two-sided p = 0.05, Student’s t test, n = 8 animals), but there was no significant difference between HFD (n = 8 animals) and RD (n = 12 animals) groups of female mice. The dFI measurements are consistent with life-shortening and neutral effects of HFD in males and female animals in the same experiment, as reported in ref. 3. Effects of 8 week-long rapamycin treatment on body weight and dFI. Body weight (e) and dFI (f) were measured every 1 and 2 weeks, respectively. All the data are presented as the mean ± SEM (n = 48 and n = 12 mice in control and rapamycin groups, respectively). g The statistics of the dFI increments in pairs of consecutive measurements are different by the presence or the absence of the treatment (the blue box, including both the control and rapamycin-treated group after cessation of the treatment, n = 204 animals, vs. the treated group, the orange box, n = 36 animals). Boxplots indicate median, 25th and 75th percentiles, whiskers indicate 5th and 95th percentiles. The dFI increments between the subsequent measurements were significantly lower under treatment (p = 0.05, Student’s two-tailed t test), thus suggesting lifespan increasing effects of rapamycin.
The correlation of this SC proxy (total luciferase flux) with dFI was even stronger (r = 0.69, p < 0.001; Fig. 6b). The observation fits well with our theoretical prediction: the kinetic equation for the order parameter (dFI) is autonomous (its coefficients do not depend on the time). This means that the organism state variables, including measures of senescent cell burden, may not explicitly depend on age, so the age dependence may only come through the dependence on dFI.
To assess the utility of dFI for the analysis of effects of interventions on aging, we retrospectively evaluated the dFI and its relation to the remaining lifespan in response to a high-fat diet (HFD) using the data from the earlier experiment³. Male mice were fed HFD instead of a regular diet (RD) beginning at 50 weeks of age had significantly reduced lifespans (Supplementary Fig. 9a) and also showed a significant increase in average dFI measured at week 78 (p = 0.05, Student’s two-tailed t-test; Fig. 6c) in comparison to control RD-fed males. In contrast, HFD feeding of female mice had no effect on either lifespan or average dFI (Supplementary Fig. 9b and Fig. 6d). Thus, dFI appeared to be a good predictor of gender-dependent differences in the organism to HFD, the underlying reasons for which remain to be explained.
Our calculations shown that the aging acceleration captured by the dFI agreed with the previously reported gain in the PFI (p = 0.02 according to Student’s t-test) in male and no effect in female mice at week 78 in the same experiment. This is a notable statement regarding the sensitivity of dFI, since PFI was developed (trained) in the MA0071 experiment and employs more (18 vs. 12) variables than mere CBC for its determination. On the contrary, we did not use any data from the MA0071-73 datasets to train the dFI model. Figure 6c, d demonstrates dFI performance relative to the independently developed PFI in an external dataset.
In another experiment, we tested the response of dFI to a short treatment with an established life-extending agent, rapamycin^36,37. Here we present the results of an experiment with a cohort of 60 60-week-old C57BL/6 male mice. 48 animals (the treatment group) were treated with rapamycin daily at a dose of 12 mg/kg for 8 weeks. The other 12 animals (the control group) received the vehicle on the same schedule.
BWs were measured every week and increased as expected in the control group (Fig. 6e). In contrast, BW in the rapamycin-treated group stayed approximately constant near the initial value throughout the observation period of 10 weeks. A lower BW is typical for rapamycin-treated mice compared to the control group^3,38. To generate dFI values for the mice in this experiment, we collected and produced CBC from blood samples from each animal every two weeks (Fig. 6f).
The longitudinal character of sampling in the experiment let us use the autoregression analysis to detect possible effects of a drug on the dynamics of dFI and hence aging. Whenever a non-random force (that is the effect of the drug) is present in Eq. (1), the jump in dFI between any of the consequent measurements from the same animal should satisfy the modified Eq. (7):
where J is the accumulated effect of the drug along the aging trajectory between the measurements. The time intervals are usually very small, αΔt ≪ 1 and hence the autoregression coefficient r ≈ 1. We, therefore, expected to identify the effect of rapamycin by comparing the distributions of the dFI increments, ΔdFI = z(t + Δt) − z(t), in animals receiving or not the treatment between the measurements both in the treated and in the control groups (see Methods for the details of the statistical analysis). The statistical analysis of the dFI increments demonstrated a significant aging deceleration associated with the rapamycin treatment (Fig. 6g, p = 0.05, Student’s two-tailed test).
We introduced a novel way to reveal biomarkers of age and frailty from big biomedical data involving longitudinal measurements, i.e., multiple samples of the same animals collected along the aging trajectories. The approach belongs to the class of unsupervised learning algorithms such that it does not require labels associated with age, mortality, and morbidity. It could be exemplified by the discovery and characterization of a biomarker of aging in mice, the dFI, from conventional and automated measurements of CBC and trained from the data from MPD.
Aging is a very slow process that occurs at characteristic time scales far exceeding times associated with molecular processes or operations of an organism’s functional subsystems. Typically, such a hierarchy of scales arises from criticality, which is a special case of a dynamic system operating close to a tipping point separating the stable and unstable region^15,16,17,39. In ref. 18, we proposed that aging results from inherent dynamic instability of the underlying regulatory networks and manifests itself as small deviations of the organism state variables (physiological indices) get exponentially amplified and lead to the exponential acceleration of mortality. The first principal component score is then an approximation to the order parameter characterizing the unstable phase and having the meaning of the total number of regulatory errors accumulated in the course of life of the animal¹⁷. Hence, we believe that aging at criticality conjecture provides a good explanation for the success of Principal Components Analysis (PCA) as a semi-quantitative tool in aging research^40,41,42.
The idea of the order parameter associated with instability is a generalization of a concept initially introduced in the Ginzburg-Landau theory to describe phase transitions in thermodynamics. The idea was further developed for applications to open non-equilibrium systems in the form of the “enslaving-principle”¹³, which states that next to the critical point, the dynamics of fast-relaxing (stable) components of a system is completely determined by the ’slow’ dynamics of only a few ’order-parameters’. The dFI identified as an approximation to the order parameter is then not a mere machine learning tool developed for specific predictions but rather a fundamental macroscopic property of the aging organism as a non-equilibrium system.
However, the abilities of linear rank reduction techniques, such as PCA, to unravel an accurate dynamic description of aging are limited for the following reasons. First, there are no reasons to believe that the effects of non-linear interactions between different dynamic subsystems are small. That is why a biomarker produced from such a linear analysis cannot be expected to perform well in different biological contexts (strains, laboratory conditions, or therapeutic interventions such as drugs). Second, biological measurements are often noisy. Hence, simple techniques lacking efficient regularization may fail to reconstruct the space of the latent variables correctly unless a prohibitively large number of samples is obtained⁴³. Finally, the association of the first principal component with the order parameter is only an approximate statement. Fundamentally, there is no way to identify the system’s dynamics from the data, which does not include the dynamics itself in multiple measurements of the same organism along the aging trajectory.
To compensate for the drawbacks of PCA and demonstrate an efficient way to produce an accurate approximation to dFI from the data, we employed an artificial neural network, a combination of a deep denoising AE and an AR model. The AE part of the algorithm is a non-linear generalization of PCA and was used to compress the correlated and necessarily noisy biological measurements into a compact set of latent variables, a low-dimensional representation of the organism state. The AR-arm of the network is nothing else but the best possible prediction of the future state of the same animal from the current measurements. The solution to the AR-problem provided by the network is the estimate for the order parameter from the given data, the dynamic frailty indicator (dFI).
The network architecture applied here was inspired by deep rank-reduction architectures, recently used to characterize numerical solutions of large non-linear dynamical systems^44,45. In a broader sense, we report the development and characterization of a fully interpretable (model-based) algorithm, a particular example of identifying the equations of motion of the underlying dynamical system from data with the help of deep learning^46,47. By the standard of physical sciences, aging in mice turns out to be “exactly solvable”. We demonstrated an analytically tractable model tying the dynamics of the physiological state and all-cause survival to the evolution of a single age-dependent quantity, the order parameter of the unstable phase, and following a simple first-order stochastic differential equation.
The profound correlations between the CBC features and dFI may reflect a key role in aging of hematopoietic tissue in determining aging of the whole organism. This concept is intuitively acceptable given the universal systemic physiological function of blood. As an alternative explanation, age-dependent changes in blood parameters may be secondary events induced by aging of the remainder of the organism (i.e., various solid tissues). However, accumulated experimental evidence argues against this. Multiple reports are demonstrating “rejuvenating” effects of young hematopoietic system on old animals delivered either by bone marrow transplantation or by parabiosis (reviewed in ref. 48). Moreover, restoration of mouse hematopoiesis through transplantation of hematopoietic stem cells (HSCs) from young vs. old donors clearly demonstrated that aged HSCs cannot be rejuvenated by the environment of a young body⁴⁹. Also, the interpretation of age dependence of HSC-derived features as secondary effects of aging would face formal difficulties since the dynamics of such factors should exhibit shorter, in fact at least twice shorter, doubling times than the dFI and the mortality rate doubling times.
A peculiar result of our analysis is that our data strongly point toward myeloid lineage which provides much more accurate predictors of biological age than lymphoid lineage parameters. This is counterintuitive since aging is generally accepted to be associated with the well-documented general decline in immunity known as an immunosenescence⁵⁰, the phenomenon illustrated by the reduced efficiency of vaccination⁵¹ and increased frequency and lethality of infectious diseases and cancer in older organisms⁵². Nevertheless, there is strong experimental evidence that supports and provides a mechanistic explanation for our finding that myeloid parameters weigh more heavily than lymphoid ones as biological age indicators. In a comprehensive study of the epigenetic mechanisms of HSC aging, Beerman et al.⁴⁹ described age-dependent epigenetic reprogramming that leads to a significant shift towards myeloid lineage differentiation of the progeny of aged HSCs^49,53. This shift is driven by specific changes in methylation of the DNA of HSCs that occur during mouse aging. Surprisingly, these changes in methylation, which alter gene expression, do not occur in the part of the genome that controls HSC phenotype, but rather modify DNA regions encoding genes that control downstream differentiation stages. Remarkably, the pattern of DNA methylation changes associated with aging of HSCs seems to represent the same process that was previously described as a DNA methylation-based clock^1,49, and therefore, may be part of the same epigenetically controlled fundamental aging mechanism. Another factor that could diminish the impact of lymphoid lineage-related parameters as biological age markers is the reactive nature of this branch of hematopoiesis, which serves to rapidly respond to sporadic events such as viral or bacterial infection, wounding, and other types of stress requiring an emergency response usually in the form of acute inflammation. Since the time of occurrence of such events is unpredictable, age-associated changes may be masked by the noise coming from large-scale age-unrelated fluctuations in the lymphoid compartment.
These observations do not mean that the blood is the single determinant of aging (otherwise, biological age would be 100% defined by the age of HSCs), but at least place it among the major drivers of the process and provide an explanation for our success in reliably determining biological age from blood test data. Aging in our model emerges cooperatively so that despite the appearance of “aging clock” in the form of the order parameter or its estimate from the data, the dFI, there is no specific subsystem tracking time (or the age) in an animal.
The organism state’s dynamics are described by an autonomous Eq. (1), where none of the model coefficients explicitly depend on time. Accordingly, no physiological state variable can depend on the age of the animals explicitly, only implicitly via dependence on the collective variable, the dFI. That is why we observed the correlations between the dFI, on the one hand, and multiple aging and frailty measures commonly known as the hallmarks of aging⁵⁴, on the other. These include grip strength, BW, RDW, and markers of inflammation such as CRP and KC (IL-8). dFI also correlated well with the p16-luciferase flux, a proxy for the number of senescent cells in aged mice. Conversely, we predict that anti-aging treatment altering the dFI should significantly alleviate other hallmarks of aging at the same time. This property is not uncommon among interventions with potentially life-extending effect⁵⁵.
Moreover, the instability of the organism state means that the animals cannot relax to any equilibrium even after a short perturbation (formally, this property manifests itself as strong auto-correlations of dFI over extended periods of time). Therefore, the effects of short treatments should likely persist until the end of life. Accordingly, the effects of such transient treatments could, in principle, be detected in short experiments involving longitudinal dFI measurements in just over a few weeks. Although the effect size and the signal-to-noise ratio in the short prospective rapamycin treatment experiment are not very large, the aging deceleration observed in this work is compatible with the earlier reported life-long pro-longevity effects of transient rapamycin treatments³⁷.
The common biomarkers of aging rely on individual or composite measures of the deviation of an organism from its youthful state. Notable examples are forms of frailty index (FI) defined as the proportion of accumulated deficits in any available signal^8,9 or, more specifically in blood test data^{9,10,11,56,57}. Since we do not expect any dependence of the physiological state variables on age other than the dependence on the order parameter, there should be a good concordance between composite markers of aging and dFI. Indeed, we observed a very high degree of concordance between the dFI and the PFI from ref. 3, which is an example of the Mahalanobis distance from old to young states⁷ and is derived from a wide range of features including CBC, physical fitness, cardiovascular health, and biochemistry.
Of all such measures, dFI alone comes with an established equation of motion (1) and hence has the best auto-correlation property along the individual aging trajectories (see, e.g., the auto-correlation of dFI and the first PC in Fig. 2 and Supplementary Fig. 2, respectively). The property is convenient for determining the effects of drugs or other interventions on aging in experiments involving comparisons of the organism state of the same animals before and during the treatments.
The reduction of aging to the dynamics of a single variable makes the quantitative model of aging in the form of the stochastic Eq. (1) from ref. 18 and this study distinct from previous proposals to derive mortality from a physiological state’s dynamics^58,59. Such models are mathematical metaphors of random walks in very high dimensional spaces and thus may be difficult to interpret or infer from biological signals without additional assumptions. We believe that criticality is a helpful theoretical observation and deep learning is a great practical tool, which may be used together to simplify identifying the model parameters and their relations to organism properties. We note that the explanation of the Gompertz law of mortality is insufficient proof of an aging theory. Matching stochastic longitudinal dynamics of physiological indices to a model prediction is a much harder challenge and should be used as an effective tool for model validation.
dFI increased exponentially with age at a characteristic doubling rate of 0.02 per week. This estimate is somewhat smaller than (but still of the same order as) the expected Gompertz mortality acceleration rate of 0.037 per week²² for the SWR/J strain. Given the observed dFI doubling rates in mice, (alpha bar{t} sim 3) and hence the deterministic phenotypic changes dominate the random effects by a factor of 3 (see Eq. (2) and Fig. 3).
More interestingly, in the cross-sectional dataset, the dFI saturated at a limiting value, which is reached at the age corresponding to the average lifespan in the group. However, we observed that the dFI ceiling corresponds to the dFI levels in cohorts of animals scheduled for euthanasia due to excessive morbidity under current laboratory protocols, which is as close to death from natural causes as animals could possibly be in a modern laboratory.
Both features of the aging trajectories in MPD are compatible with the analytical solutions of Eq. (1) for the dynamics of the order parameter. In ref. 18, we explained that early in life dFI increases exponentially (see Eq. (2)). At the age approximately corresponding to the average lifespan in the population, non-linear effects take over the dynamics of dFI, and the organism state deviates from its youthful state even faster than exponentially. Such a situation is incompatible with survival and hence cannot be observed in the data. In our model and in the experiment, death occurs quickly once the maximum dFI level is reached at some point in the life history of the animal.
The stochastic Eq. (1) establishes the “law of motion” for the organism’s physiological state and predicts the late-life mortality deceleration in the form of saturation of mortality at the plateau level, (M(t, gg ,bar{t}), approx ,alpha). The predicted relationship between the limiting mortality and the mortality rate doubling time held in a variety of species⁶⁰ and experiments in multiple conditions in the same species, such as nematodes⁴². Here we report the validation of the limiting mortality prediction in very large cohorts of mice from ref. 25.
The good semi-quantitative agreement between the empirical mortality curves in large experiments and the theoretical prediction provides an independent and sensitive test of the aging at criticality model as a theoretical framework proposed here for the data analysis in experiments involving aging animals. More specifically, the experimental confirmation for the late-life mortality deceleration prediction validates the basic stochastic Eq. (1) and the association between its solution in the form of dFI and all-cause mortality. We also note that the mortality deceleration in the model arises from the stochastic nature of the order parameter dynamics and should be expected even in a cohort of genetically identical animals (see the discussion in ref. 61).
Deviations from the Gompertz law in human cohorts also occur, but at ages exceeding the average lifespan when the mortality is already well beyond the theoretical limit corresponding to the mortality rate doubling rate by almost an order of magnitude⁶⁰. This means that the character of the organism state dynamics in the course of human aging is qualitatively different than that in mice or nematodes. In ref. 5, we observed that the fluctuations of physiological indices in humans are also dominated by a collective variable characterized by a relatively long but finite auto-correlation time (in the range of a few weeks) and associated with age and all-cause mortality. The number of individuals exhibiting signs of the loss of dynamic stability (measured by exceedingly long auto-correlation times) increased exponentially with age at a rate matching the mortality doubling rate from the Gompertz mortality law⁶¹.
The intimate relation between the auto-correlation properties of physiological state variables and hallmarks of aging suggests that AR analysis enhanced by deep learning may help discover signatures of human aging and chronic disease progression. The benefit may be particularly huge in studies involving large sets of longitudinal measurements but often lacking follow-up mortality and morbidity information. While hallmarks of aging in mice are correlated and primarily reversible, a large part of physiological changes associated with aging in humans is stochastic and may be thermodynamically irreversible⁶². Therefore, we expect that the systematic application of dynamic systems theory principles to biomedical data analysis will help identify actionable aging phenotypes and thus facilitate the discovery and development of anti-aging therapeutics that produce lasting rejuvenating effects.
The training data set was prepared from the nine data sources available in the MPD¹⁹. A list of the included sources is presented in Supplementary Data 1 with references to the included and missing records grouped by sex and age cohorts. We used the assays providing the CBC data only, assays with other biomarkers were not considered due to the insufficient number of samples. Our model was trained using the best overlap of available CBC features from all sources. The final list contained 12 CBC features: granulocytes differential (GR%), granulocytes count (GR), hemoglobin (HB), hematocrit (HCT%), lymphocyte differential (LY%), lymphocyte count (LY), mean corpuscular hemoglobin content (MCHC), mean hemoglobin concentration (MCH), mean corpuscular volume (MCV), platelet count (PLT), red blood cell count (RBC) and white blood cell count (WBC). See Supplementary Data 2 for the list of all abbreviations. If the data source lacked granulocytes measurements, it was retrieved using formulas:
All animals with one or more missing parameters were excluded from the training. The percentage of the excluded records was <2% and should not have affected the results.
All animal experiments were approved by the Institutional Animal Care and Use Committee of Roswell Park Cancer Institute or by the Explora BioLabs, Inc Animal Use Committee.
We received 4–5-week-old NIH Swiss male and female mice from Charles River Laboratories (Wilmington, MA). They were allowed to age within the Roswell Park Comprehensive Cancer Center (RPCCC) animal facility. During this time mice were housed 1–3 per cage and were fed ad lib with standard chow (Tekland Global 18% Protein Rodent Diet). Blood samples were obtained at different ages as part of creating of the PFI³. Blood samples were collected from a single submandibular vein bleed into EDTA-treated Vacutainer tubes (total volume of 20 μl) and used for whole blood cell counts and glucose measurements using Hemavet 950 Analyzer (Drew Scientific). Another 75 μl of blood was collected into Li-Heparin treated plasma separator tubes; plasma was purified by centrifugation at 5000×g for 5 min and used for measuring the concentration of circulating pro-inflammatory cytokines and triglycerides.
Dataset MA0071 was built in a cross-sectional experiment using male and female NIH Swiss mice. Blood was collected from male mice at ages of 26 (n = 20), 64 (n = 18), 78 (n = 17), 92 (n = 14), and 132 (n = 6) weeks. Female age groups were represented by the ages of 30 (n = 20), 56 (n = 20), 68 (n = 19), 82 (n = 19), 95 (n = 18), 108 (n = 20), and 136 (n = 7) weeks. Dataset MA0072 was obtained from a longitudinal experiment. Blood samples were collected at the ages of 66 (n = 27), 81 (n = 22), 94 (n = 21), 109 (n = 16), and 130 (n = 7) weeks. Dataset MA0073 includes blood samples collected from 97 male and 127 female mice of different ages when animals reached approved experimental endpoints and required humane euthanasia. All animal procedures were performed according to the approved Institutional Animal Care and Use Committee (IACUC) protocol. Mice were monitored daily for the development of age-related pathologies. Whenever health issues were reported, research staff contacted veterinarian staff and followed their recommendations for either treatment or euthanasia. Mice were treated until the condition was improved or euthanized when the endpoint for each health condition described in the protocol was reached. Euthanasia was performed by CO2 asphyxiation followed by cervical dislocation. See Supplementary Data 3 for the total number of animals in these datasets.
p16/INK4a-LUC female mice (p16-Luc) at ages of 44 to 106 weeks were obtained from the N. Sharpless laboratory at the University of North Carolina (Chapel Hill, NC). All animals were housed under 12:12 light:dark conditions (12 hours of light followed by 12 hours of darkness) at the Laboratory Animal Shared Resource at RPCCC. All animal experiments were approved by the IACUC of Roswell Park Cancer Institute. Bioluminescence imaging was performed using an IVIS Spectrum imaging system (Caliper LifeSciences, Inc, Waltham, MA). p16/Ink4a-Luc+/- mice were injected intraperitoneally with D-Luciferin (150 mg/kg, Gold Biotechnology), 3 minutes later anesthetized with isoflurane and imaged using a 20-second integration time and medium binning. The images were processed and quantified as the sum of photon flux recorded from both sides of each mouse using Living Image software (Perkin Elmer, Waltham, MA.).
For the rapamycin treatment experiment 60-week-old C57BL/6J male mice were obtained from Jackson Laboratories (USA). The cohort of 60 60-week-old C57BL/6 male mice was divided into treatment (n = 12) and control (n = 48) groups using a stratified randomization technique to produce indistinguishable distributions of dFI values prior to the experiment. The blood samples (total volume of 120 μL) were collected into EDTA tubes via submandibular or facial vein using a lancet. All animal procedures were approved by the Explora BioLabs, Inc animal use committee (IACUC SP17-004-035B) and were in accordance with Explora BioLabs, Inc policies on the care, welfare, and treatment of laboratory animals. Rapamycin was purchased from LC Laboratories (MA, USA). Rapamycin was administered daily at 12 mg/kg via oral gavage for 8 weeks. The control group was treated with vehicle (5% Tween-80, 5% PEG-400, 3% DMSO).
We performed the PCA with the help of Python and Scikit-learn package⁶³. First, we applied PCA transformation to the entire training dataset. However, the principal components were dominated by the difference in mice strains. We removed strain difference by subtracting mean values of CBC features calculated for the earliest age available for the selected strain from values of CBC features of all animals for this strain:
where the indices i and j enumerate the CBC features and strains, respectively, and t is the age of an animal. For simplicity, we filtered out mouse strains, which were not presented in the Peters4 dataset⁶⁴.
Most of the variance in the data representing the full dataset was associated with animal growth and maturation. The first PC score increased with the age of animals most notably after 25 weeks (Supplementary Fig. 1). On the contrary, the second and the third PC scores acquire non-zero means by the same age of 25 weeks. This suggests that aging and early development in mice are different phenotypes. Subsequently, we performed all our calculations using the data from animals older than 25 weeks.
The death records for animals linked with the MPD dataset Peters4 were also available in the MPD as a separate dataset named Yuan2²³. These datasets contain different cohorts of animals with a vast overlap. We found mortality records for 487 animals in Peters4 dataset, while 393 animals were missing. The reason for the missingness is unknown. To create censoring records, we included all animals having at least two sequential CBC measurements. We assumed that the animals provided with a single CBC measurement were probably sacrificed after the blood collection. Hence, if an animal has more than one CBC measurement, we consider it lost at the last measurement. Altogether, we found 79 animals satisfying this condition. The rest 314 animals with unknown mortality dates were excluded from the analysis.
The Spearman’s rank correlation test was performed separately for the two cohorts of mice. The first cohort included all animals from the Peters4 dataset with uncensored mortality records. The second cohort included animals from the Peters4 dataset with the measurements of BW and the IGF1 serum level taken from the MPD dataset named Yuan1²⁴.
We performed the Cox PH regression analysis with the help of Python and Lifelines package⁶⁵. First, we produced the supervised multivariate Cox-PH model using the age, sex, and CBC features as covariates and utilizing the data from all the animals from the Peters4 dataset, including both censored and uncensored mortality events. The output of the model, the log-hazard ratio (HR_CBC) is the dot product of the vector comprising the CBC features and the respective coefficients from the Cox-PH model. Next, we tested the association of the HR_CBC feature in a univariate Cox-PH model alongside the dFI score in cohorts of animals of the same age and sex (see Supplementary Data 6) as an alternative for Spearman’s rank correlation test. Both tests were in good agreement with each other.
We used a combination of a deep AE and a simple AR model for modal analysis (AE-AR model). At its bottleneck, the encoder arm of the AE produced a compressed 4-dimensional representation y of the input, the 12-dimensional physiological state vectors x built from the available CBC measurements. The decoder arm reconstructed the original 12-dimensional state (tilde{{{{{{{{bf{x}}}}}}}}}) from the bottleneck features.
Simultaneously with the AE, we trained the network to fit the longitudinal slice of MPD (including fully-grown animals at ages from 26 to 104 weeks with a sampling interval of Δt = 26 weeks) to the solution of the linearized (g = 0) version of Eq. (1),
where z is the best possible linear combination of AE bottle-neck features, (r=exp (alpha Delta t), approx ,1), and ({z}^{prime}) are the best fit values of the auto-regression coefficient and the constant shift, respectively. Here, ξ is the error of the fit (the combination of the system’s noise and measurement errors).
We adapted the neural network architecture proposed in ref. 45. Our implementation handled cross-sectional and longitudinal measurements simultaneously with an imbalance in favor of cross-sectional data, which is typical for real-world clinical data. As inputs, the network has three 12-dimensional vectors represented by CBC parameters: one for the cross-sectional dataset (x), and two others for the longitudinal dataset corresponding to the present (x_n) and future (x_n+1) states of a sample (Fig. 7a).
a The network includes of nonlinear auto-encoder (AE) and the auto-regression model (AR), the linear projector, and the decoder blocks. The output of the projector block was designated as dFI. b Schematic representations of the residual network (ResNet) blocks.
The cross-sectional samples served as inputs in the AE. The longitudinal samples in the compressed representation (y_n, y_n+1) also participated in the training autoregression part. The main advantage of the inclusion of AE in the neural network is its ability of effective nonlinear dimensionality reduction⁶⁶, which is necessary for such correlated quantities as CBC components (Fig. 1a). The reduced size of latent dimensions works as regularization and helps to train without overfitting on a small longitudinal dataset using more samples from a larger cross-sectional dataset.
The AE block encodes input (x) to the 4-dimensional vector y = ϕ(x) and then reconstructs back the original signal (widetilde{{{{{{{{bf{x}}}}}}}}}={phi }^{-1}({{{{{{{bf{y}}}}}}}})). AE was implemented as a stack of fully connected dense layers and residual network blocks (ResNet)⁶⁷. The dense layers have a trainable weight matrix W, bias vector b, and a linear activation function by default. The ResNet block, shown in Fig. 7b, is a stack of two dense layers with an activation function of a leaky rectified linear unit (Leaky ReLU). The input and the output are linked by applying element-wise addition. The ResNet blocks add nonlinear rectification transformations to the original input, helping to learn nonlinear transformations. The AE is trained simultaneously on cross-sectional and longitudinal datasets.
The projector block takes a 4-dimensional vector as an input and transforms it to a scalar z = A ⋅ y, which we refer to as dFI. During training, a pair of vectors is fed to the inputs: one y_n for the present state of the system and one y_n+1 for the future state. The linear dynamics block solves the autoregression problem (7) and predicts the future state ({widetilde{z}}_{n+1}=xi ({z}_{n})=r{z}_{n}+b). The auxiliary decoder block reconstructs the original 12-dimension CBC vector from the output of the linear dynamics block ({widetilde{z}}_{n+1}) utilizing the decoder ϕ⁻¹ from the AE block: ({widetilde{{{{{{{{bf{x}}}}}}}}}}_{n+1}={phi }^{-1}({{{{{{{bf{B}}}}}}}}cdot {widetilde{{{{{{{{bf{z}}}}}}}}}}_{n+1})).
To force matrices of A and B in the projector and linear dynamics blocks to be left and right eigenvectors in the solution of Eq. (7) we added the following constraints:
The total loss function is the weighted sum of the following losses:
where
Here, ({{{{{{{{mathcal{L}}}}}}}}}_{{{{{{{{rm{AE}}}}}}}}}) is the AE reconstruction loss, ({{{{{{{{mathcal{L}}}}}}}}}_{{{{{{{{rm{pred}}}}}}}}}) is the future state reconstruction loss, ({{{{{{{{mathcal{L}}}}}}}}}_{{{{{{{{rm{AR}}}}}}}}}) the auto-regression loss, ({{{{{{{{mathcal{L}}}}}}}}}_{{{{{{{{rm{C}}}}}}}}}) is the loss to force the constraints from Eq. (8), and the term (parallel {{{{{{{bf{W}}}}}}}}{parallel }_{2}^{2}) is L2 regularization of NN weights to avoid over-fitting issue.
The weights α₁, α₃, α₄ were assigned to the values of 1, 100, and 0.01, respectively. The weight α₂ was gradually increased from 0 to 1 during training. The model was trained for 600 epochs with a learning rate of 0.001 and Adam optimizer⁶⁸. The last 200 epochs were trained with a learning rate of 0.0001. The AE/AR NN architecture was implemented with Python and TensorFlow framework⁶⁹.
The non-linear dynamics of the order parameter are crucial for explaining mortality. At the same time, the effects of the nonlinearity can be neglected almost always in the course of the life of an animal if the dimensionless parameter expressing the animal lifespan (bar{t}) in units of the mortality rate doubling time is large, (alpha bar{t}, gg ,1)¹⁸. Given the observed dFI doubling rates in mice, (alpha bar{t} sim 3) and hence the linear AR model is only a reasonable approximation. One should obtain better dFI variants in the future by increasing the rank in AR models, possibly including the effects of mode coupling with dFI.
The model was validated in test datasets (see Supplementary Data 3), which were completely excluded from the training of the AE-AR model. The test datasets were obtained from independent experiments by collecting CBC samples from cohorts of NIH Swiss mice of different age and sex (dataset MA0071), cohort NIH Swiss male mice observed for 15 months (dataset MA0072), and cohorts of naive male and female NIH Swiss mice that were humanely euthanized after reaching approved experimental endpoints (dataset MA0073).
We estimated the reconstruction error of the AE by calculation of the root mean squared error (RMSE) and the coefficient of determination R² for each CBC feature in training and test sets (Supplementary Data 4 and Supplementary Data 5). The average RMSE in the test set was 229.6 with R² = 0.55; in the training set, RMSE was 106.4 and R² = 0.77. The best reconstruction was achieved for hematocrit (R² = 0.95), red blood cells (R² = 0.92) and lymphocytes (R² = 0.87); the worst results were for mean corpuscular hemoglobin concentration (R² = − 0.82) and platelets (R² = − 0.14) in the test set. We note that according to definition, −1 < R² < 1 (see, e.g.,⁷⁰), the quantity may be negative in either the train or validation sets, which indicates cases of particularly bad fit.
We performed the investigation of the effects of rapamycin on dFI trajectories in individual animals. Technically, we compared the increments of dFI levels along the individual life histories in time intervals depending on the amount of treatment between the subsequent time points. Such analysis explicitly relies on the equation of motion (7) for the order parameter, which is approximated by dFI. The drug’s effect manifests itself as the “force” term reducing the dFI increments between the measurements when the drug is given and having no effect (no force) whenever the drug is not administered, both in the treatment and control groups.
The determination of a drug’s effect on the aging process is therefore equivalent to determining the “force” term in the autoregression problem in Eq. (4). Since the natural variation of dFI levels between animals is often high, longitudinal studies should have more statistical power than standard group comparisons. The dFI was trained with the AR model (7). Accordingly, it is well suited to maximize the signal-to-noise ratio in a longitudinal analysis of an anti-aging intervention’s effects. If required, the autoregression model can accommodate any number of confounding factors, such as the experimental batch or sex of the animals. Technically, one can achieve the goal by adding the respective covariates to Eq’s right-hand side (4).
The data for mice mortality were taken from ref. 25. Only control groups were selected for the analysis. Mice removed from the study were also removed from the current survival analysis. The mice were pooled together from all three study centers and cohorts and separated into two groups according to sex. Overall, there were 3249 male mice, and 2978 female mice.
The mortality analysis is done with the help of the Nelson-Aalen fitter from the lifelines python package⁶⁵. For the Gompertz fit of the survivals and other survival analysis, we used the custom code published on GitHub.
Further information on research design is available in the Nature Research Reporting Summary linked to this article.
The data supporting the findings of this study are available at the MPD (RRID:SCR_003212). Raw data files and scripts to reproduce all findings are available on GitHub https://github.com/gero-science/mice_dfi⁷¹. Additional data are available from the corresponding authors on reasonable request. Source data are provided with this paper.
The code will be available on GitHub https://github.com/gero-science/mice_dfi⁷¹
Horvath, S. DNA methylation age of human tissues and cell types. Genome Biol. 14, R115 (2013).
Article  PubMed  PubMed Central  Google Scholar 
Levine, M. E. et al. An epigenetic biomarker of aging for lifespan and healthspan. Aging 10, 573 (2018).
Article  PubMed  PubMed Central  Google Scholar 
Antoch, M. P. et al. Physiological frailty index (pfi): quantitative in-life estimate of individual biological age in mice. Aging 9, 615 (2017).
Article  PubMed  PubMed Central  CAS  Google Scholar 
Putin, E. et al. Deep biomarkers of human aging: application of deep neural networks to biomarker development. Aging 8, 1021 (2016).
Article  PubMed  PubMed Central  CAS  Google Scholar 
Pyrkov, T. V. et al. Longitudinal analysis of blood markers reveals progressive loss of resilience and predicts ultimate limit of human lifespan. Nat. Commun. 12, 2765 (2019).
Schultz, M. B. et al. Age and life expectancy clocks based on machine learning analysis of mouse frailty. Nat. Commun. 11, 1 (2020).
Google Scholar 
Cohen, A. A. et al. A novel statistical approach shows evidence for multi-system physiological dysregulation during aging. Mechanisms Ageing Dev. 134, 110 (2013).
Article  Google Scholar 
Mitnitski, A. B., Mogilner, A. J. & Rockwood, K. Accumulation of deficits as a proxy measure of aging. ScientificWorldJournal 1, 323–336 (2001).
Mitnitski, A. et al. Age-related frailty and its association with biological markers of ageing. BMC Med. 13, 161 (2015).
Article  PubMed  PubMed Central  Google Scholar 
Blodgett, J. M., Theou, O., Mitnitski, A., Howlett, S. E. & Rockwood, K. Associations between a laboratory frailty index and adverse health outcomes across age and sex. Aging Med. 2, 11 (2019).
Article  Google Scholar 
Justice, J. et al. Senolytics in idiopathic pulmonary fibrosis: results from a first-in-human, open-label, pilot study. ebiomedicine 40, 554–563 (2019).
Article  PubMed  PubMed Central  Google Scholar 
Fahy, G. M. et al. Reversal of epigenetic aging and immunosenescent trends in humans. Aging Cell 18, e13028 (2019).
Article  PubMed  PubMed Central  CAS  Google Scholar 
Hermann Haken. Phys. Astron. online Libr. https://doi.org/10.1007/978-3-662-10184-1 (Springer, Berlin, Heidelberg, 2004).
Scheffer, M. et al. Early-warning signals for critical transitions. Nature 461, 53 (2009).
Article  ADS  PubMed  CAS  Google Scholar 
Scheffer, M. et al. Quantifying resilience of humans and other animals. Proc. Natl Acad. Sci. USA 115, 11883 (2018).
Article  ADS  PubMed  PubMed Central  CAS  Google Scholar 
Krotov, D., Dubuis, J. O., Gregor, T. & Bialek, W. Morphogenesis at criticality. Proc. Natl Acad. Sci. 111, 3683 (2014).
Article  ADS  PubMed  PubMed Central  CAS  Google Scholar 
Kogan, V., Molodtsov, I., Menshikov, L. I., Reis, R. J. S. & Fedichev, P. Stability analysis of a model gene network links aging, stress resistance, and negligible senescence. Sci. Rep. 5, 13589 (2015).
Article  ADS  PubMed  PubMed Central  CAS  Google Scholar 
Podolskiy, D. Critical dynamics of gene networks is a mechanism behind ageing and Gompertz law. arXiv https://arxiv.org/abs/1502.04307 (2015).
Bogue, M. A. et al. Accessing data resources in the mouse phenome database for genetic analysis of murine life span and health span. J. Gerontology. Ser. A 71, 170 (2016).
Article  CAS  Google Scholar 
Fiedler, B. E. Handbook of Dynamical Systems, Vol. 2 (Gulf Professional Publishing, 2002).
Seydel, R. Practical Bifurcation and Stability Analysis, Vol. 1, p. 477 (Springer Science & Business Media, 2009).
Hughes, B. G. & Hekimi, S. Different mechanisms of longevity in long-lived mouse and Caenorhabditis elegans mutants revealed by statistical analysis of mortality rates. Genetics 204, 905 (2016).
Article  PubMed  PubMed Central  Google Scholar 
Yuan, R. et al. Genetic coregulation of age of female sexual maturation and lifespan through circulating igf1 among inbred mouse strains. Proc. Natl Acad. Sci. 109, 8224 (2012).
Article  ADS  PubMed  PubMed Central  CAS  Google Scholar 
Yuan, R. et al. Aging in inbred strains of mice: study design and interim report on median lifespans and circulating igf1 levels. Aging Cell 8, 277 (2009).
Article  PubMed  CAS  Google Scholar 
Harrison, D. E. et al. Rapamycin fed late in life extends lifespan in genetically heterogeneous mice. Nature 460, 392 (2009).
Article  ADS  PubMed  PubMed Central  CAS  Google Scholar 
O’Connell, K. E. et al. Practical murine hematopathology: a comparative review and implications for research. Comp. Med. 65, 96 (2015).
PubMed  PubMed Central  Google Scholar 
Patel, K. V. et al. Red Cell distribution width and mortality in older adults: a meta-analysis. J. Gerontol. 65 A, 258 (2010).
Article  Google Scholar 
Baggiolini, M. Chemokines and leukocyte traffic. https://doi.org/10.1038/33340 (1998).
Harris, T. B. et al. Associations of elevated interleukin-6 and C-reactive protein levels with mortality in the elderly. Am. J. Med. 106, 506 (1999).
Article  PubMed  CAS  Google Scholar 
Mahmoudi, S., Xu, L. & Brunet, A. Turning back time with emerging rejuvenation strategies. Nat. Cell Biol. 21, 32 (2019).
Article  PubMed  PubMed Central  CAS  Google Scholar 
Kuilman, T., Michaloglou, C., Mooi, W. J. & Peeper, D. S. The essence of senescence. Genes Dev. 24, 2463 (2010).
Article  PubMed  PubMed Central  CAS  Google Scholar 
van Deursen, J. M. The role of senescent cells in ageing. Nature 509, 439 (2014).
Article  ADS  PubMed  PubMed Central  Google Scholar 
Hall, B. M. et al. Aging of mice is associated with p16(Ink4a)- and β-galactosidasepositive macrophage accumulation that can be induced in young mice by senescent cells. Aging 8, 1294 (2016).
Article  PubMed  PubMed Central  CAS  Google Scholar 
Kim, W. Y. & Sharpless, N. E. The regulation of INK4/ARF in cancer and aging. Cell 127, 265 (2006).
Article  PubMed  CAS  Google Scholar 
Burd, C. E. et al. Monitoring tumorigenesis and senescence in vivo with a p16 INK4a-luciferase model. Cell 152, 340 (2013).
Article  PubMed  PubMed Central  CAS  Google Scholar 
Wilkinson, J. E. et al. Rapamycin slows aging in mice. Aging Cell 11, 675 (2012).
Article  PubMed  CAS  Google Scholar 
Bitto, A. et al. Transient rapamycin treatment can increase lifespan and healthspan in middle-aged mice. elife 5, e16351 (2016).
Article  PubMed  PubMed Central  Google Scholar 
Miller, R. A. et al. Rapamycin-mediated lifespan increase in mice is dose and sex dependent and metabolically distinct from dietary restriction. Aging Cell 13, 468 (2014).
Article  PubMed  PubMed Central  CAS  Google Scholar 
Balleza, E. et al. Critical dynamics in genetic regulatory networks: examples from four kingdoms. PLoS One 3, e2456 (2008).
Article  ADS  PubMed  PubMed Central  Google Scholar 
Nakamura, E., Miyao, K. & Ozeki, T. Assessment of biological age by principal component analysis. Mech. Ageing Dev. 46, 1 (1988).
Article  PubMed  CAS  Google Scholar 
Park, J., Cho, B., Kwon, H. & Lee, C. Developing a biological age assessment equation using principal component analysis and clinical biomarkers of aging in korean men. Arch. Gerontol. geriatrics 49, 7 (2009).
Article  Google Scholar 
Tarkhov, A. E. et al. A universal transcriptomic signature of age reveals the temporal scaling of caenorhabditis elegans aging trajectories. Sci. Rep. 9, 1 (2019).
Article  CAS  Google Scholar 
Johnstone, I. M. & Lu, A. Y. On consistency and sparsity for principal components analysis in high dimensions. J. Am. Stat. Assoc. 104, 682 (2009).
Article  MathSciNet  PubMed  PubMed Central  MATH  CAS  Google Scholar 
Mardt, A., Pasquali, L., Wu, H. & Noé, F. VAMPnets for deep learning of molecular kinetics. Nat. Commun. 9, 5 (2018).
Article  ADS  PubMed  PubMed Central  Google Scholar 
Lusch, B., Kutz, J. N. & Brunton, S. L. Deep learning for universal linear embeddings of nonlinear dynamics. Nat. Commun. 9, 4950 (2018).
Article  ADS  PubMed  PubMed Central  Google Scholar 
Wu, T. & Tegmark, M. Toward an artificial intelligence physicist for unsupervised learning. Phys. Rev. E 100, 033311 (2019).
Article  ADS  PubMed  CAS  Google Scholar 
Liu, Z. & Tegmark, M. Ai poincar’e: machine learning conservation laws from trajectories. Phys. Rev. Lett. 126, 180604 (2020).
Hofmann, B. Young blood rejuvenates old bodies: a call for reflection when moving from mice to men. Karger 1, 45 (2018).
Beerman, I. et al. Proliferation-dependent alterations of the DNA methylation landscape underlie hematopoietic stem cell aging. Cell Stem Cell 12, 413 (2013).
Article  PubMed  CAS  Google Scholar 
Franceschi, C. et al. Inflamm-aging: an evolutionary perspective on immunosenescence. Ann. N. Y. Acad. Sci. 908, 244 (2006).
Article  ADS  Google Scholar 
Pawelec, G. Immunosenescence and cancer. Biogerontology 18, 717 (2017).
Article  PubMed  CAS  Google Scholar 
Crooke, S. N., Ovsyannikova, I. G., Poland, G. A. & Kennedy, R. B. Immunosenescence and human vaccine immune responses. Immun. Ageing 16, 25 (2019).
Pang, W. W. et al. Human bone marrow hematopoietic stem cells are increased in frequency and myeloid-biased with age. Proc. Natl Acad. Sci. USA 108, 20012 (2011).
Article  ADS  PubMed  PubMed Central  CAS  Google Scholar 
López-Otín, C., Blasco, M. A., Partridge, L., Serrano, M. & Kroemer, G. The hallmarks of aging. Cell 153, 1194 (2013).
Article  PubMed  PubMed Central  Google Scholar 
Partridge, L., Fuentealba, M. & Kennedy, B. K. The quest to slow ageing through drug discovery. Nat. Rev. Drug Discov. 19, 513–532 (2020).
Wang, Y. et al. Prediction of chemotherapy adverse reactions and mortality in older patients with primary lung cancer through frailty index based on routine laboratory data. Clin. Interv. Aging 14, 1187 (2019).
Article  PubMed  PubMed Central  CAS  Google Scholar 
Kane, A. E., Keller, K. M., Heinze-Milne, S., Grandy, S. A. & Howlett, S. E. A murine frailty index based on clinical and laboratory measurements: links between frailty and pro-inflammatory cytokines differ in a sex-specific manner. J. Gerontol. Ser. A 74, 275 (2019).
Article  CAS  Google Scholar 
Sacher, G. & Trucco, E. The stochastic theory of mortality. Ann. N. Y. Acad. Sci. 96, 985 (1962).
Article  ADS  PubMed  MATH  CAS  Google Scholar 
Yashin, A. I., Manton, K. G. & Vaupel, J. W. Mortality and aging in a heterogeneous population: a stochastic process model with observed and unobserved variables. Theor. Popul. Biol. 27, 154 (1985).
Article  MathSciNet  PubMed  MATH  CAS  Google Scholar 
Vaupel, J. W. et al. Biodemographic trajectories of longevity. Science 280, 855 (1998).
Article  PubMed  CAS  Google Scholar 
Pyrkov, T. V., Sokolov, I. S. & Fedichev, P. O. Deep longitudinal phenotyping of wearable sensor data reveals independent markers of longevity, stress, and resilience. Aging 13, 7900 (2021).
Article  PubMed  PubMed Central  CAS  Google Scholar 
Tarkhov, A. E., Denisov, K. A., & Fedichev, P. O. Aging clocks, entropy, and the limits of age-reversal. bioRxiv (McGraw-Hill, 2022).
Pedregosa, F. et al. Scikit-learn: machine learning in {P}ython. J. Mach. Learn. Res. 12, 2825 (2011).
MathSciNet  MATH  Google Scholar 
Peterset, L. L. al. Large-scale, high-throughput screening for coagulation and hematologic phenotypes in mice. Physiol. Genomics.11, 185–193 (2002).
Davidson-Pilon, C. Lifelines. https://github.com/camdavidsonpilon/lifelines (2016)
Hinton, G. E. & Salakhutdinov, R. R. Reducing the dimensionality of data with neural networks. Science 313, 504 (2006).
Article  ADS  MathSciNet  PubMed  MATH  CAS  Google Scholar 
He, K., Zhang, X., Ren, S. & Sun, J. Deep residual learning for image recognition. in 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2016, pp. 770–778 (IEEE Computer Society, 2016).
Kingma, D. P. & Ba, J. L. Adam: a method for stochastic optimization. https://arxiv.org/abs/1412.6980 (2014).
Abadi, M. TensorFlow: large-scale machine learning on heterogeneous systems. https://www.tensorflow.org/ (2015).
Steel, R. G. D. & Torrie, J. H. Principles and procedures of statistics: with special reference to the biological sciences. (McGraw Hill, 1960).
Avchaciov, K. AI for unsupervised learning of aging principles from longitudinal data. https://doi.org/10.5281/zenodo.6991110 (2022).
Download references
We thank Norman Sharpless for generously providing p16-Luc reporter mice. This work was supported in part by National Cancer Institute (NCI) grant P30 CA016056 for use of Roswell Park Comprehensive Cancer Center’s Laboratory Animal and Translational Imaging Shared Resources (M.P.A. and A.V.G.) and by a contract from Genome Protection, Inc. (M.P.A. and A.V.G.). The authors are grateful to M. Kholin from Gero for useful comments and discussions.
Gero PTE. LTD., 409051, Singapore, Singapore
Konstantin Avchaciov, Andrei E. Tarkhov, Leonid I. Menshikov, Olga Burmistrova & Peter O. Fedichev
Department of Pharmacology and Therapeutics, Roswell Park Comprehensive Cancer Center, Buffalo, NY, USA
Marina P. Antoch
Genome Protection, Inc., Buffalo, NY, USA
Ekaterina L. Andrianova & Andrei V. Gudkov
Department of Cell Stress Biology, Roswell Park Comprehensive Cancer Center, Buffalo, NY, USA
Andrei V. Gudkov
You can also search for this author in PubMed Google Scholar
You can also search for this author in PubMed Google Scholar
You can also search for this author in PubMed Google Scholar
You can also search for this author in PubMed Google Scholar
You can also search for this author in PubMed Google Scholar
You can also search for this author in PubMed Google Scholar
You can also search for this author in PubMed Google Scholar
You can also search for this author in PubMed Google Scholar
M.P.A. and E.I.A. designed and carried out experiments, analyzed, interpreted data and reviewed the manuscript. K.A. and A.E.T. performed calculations and data analysis and wrote the manuscript. O.G., L.I.M, A.V.G, and P.O.F. discussed the results, wrote and reviewed the manuscript.
Correspondence to Peter O. Fedichev.
P.O.F is a shareholder of Gero PTE. LTD. A.V.G. is a member of Gero PTE. LTD. Advisory Board. K.A., A.E.T., L.I.M., O.B., and P.O.F. are employees of Gero PTE. LTD. A.V.G. is co-founder and shareholder of Genome Protection. E.I.A. is employee of Genome Protection. M.P.A. has no competing interests. The study was funded by Gero PTE. LTD.
Nature Communications thanks Konstantin Arbeev, Arnold Mitnitski, and the anonymous reviewers for their contribution to the peer review of this work.
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/.
Reprints and Permissions
Avchaciov, K., Antoch, M.P., Andrianova, E.L. et al. Unsupervised learning of aging principles from longitudinal data. Nat Commun 13, 6529 (2022). https://doi.org/10.1038/s41467-022-34051-9
Download citation
Received: 06 April 2020
Accepted: 06 October 2022
Published: 01 November 2022
DOI: https://doi.org/10.1038/s41467-022-34051-9
Anyone you share the following link with will be able to read this content:
Sorry, a shareable link is not currently available for this article.

Provided by the Springer Nature SharedIt content-sharing initiative
By submitting a comment you agree to abide by our Terms and Community Guidelines. If you find something abusive or that does not comply with our terms or guidelines please flag it as inappropriate.
Advertisement
Nature Communications (Nat Commun) ISSN 2041-1723 (online)
© 2022 Springer Nature Limited
Sign up for the Nature Briefing newsletter — what matters in science, free to your inbox daily.

source

The project aims to revitalize a neighborhood characterized by distressed WWII-era public housing.
U.S. Rep. Adam Smith joined Renton city leaders on Oct. 24 for a tour of the many projects that make up the Sunset Terrace Neighborhood transformation, which aims to add hundreds of units of affordable housing centered around public resources and services within walking distance.
The plan to modernize the neighborhood, which since the middle of the 20th century has been characterized by what Renton Mayor Armondo Pavone described as “dilapidated WWII-era housing,” is one involving multiple agencies including the Renton Housing Authority as well as private investors.
Since 2012, six affordable housing projects and three mixed-income projects have been finished or are under construction in the Sunset Area. According to the city, investments so far total $150 million, with another $230 million projected for current or planned projects. This is reportedly a projected sum for projects to be completed by the city, the Renton Housing Authority and for-profit developers in the next several years.

The Sunset Neighborhood is centered around the Sunset Neighborhood Park which used to be the site of the Sunset Terrace public housing project. The current 3.2-acre park includes large lawns for play and exercise, a splash pad water feature, as well as one of the country’s only 30-foot climbing structures.
Across from the park is the 14,500 square-foot Renton Highlands Library, managed by King County and that replaced the old Highlands Library which was less than half of the size. Just down the block, the site of the old library now serves as the Sunset Neighborhood Center. The facility offers physical and mental healthcare, dental healthcare and educational classes as part of what Renton Housing Authority Executive Director Mark Gropper described as a vision for a “service enriched” neighborhood.

The Sunset Neighborhood Center is a City of Renton and Renton Housing Authority project and on-site service providers include Health Point Clinic, Neighborhood House, Valley Service Behavioral Health Care, and St. Vincent de Paul Centro Rendu.
Completed housing projects in the neighborhood include Sunset Oaks — a 60-unit multi-family affordable housing development with studio, one bedroom and two bedroom apartments. Another finished project is Sunset Court, a Renton Housing Authority project built on the site of the former Sunset Park that includes 50 units of multi-family affordable housing including both town homes and apartments.

Housing projects under construction include the Renton Highlands Library-adjacent Sunset Terrace with 108 multi-family housing units to come and 4,000 square-feet of commercial space sitting directly across from Sunset Neighborhood Park.
On the other side of the park, is the site of what is being called the Solara Master Plan. It is a private redevelopment of the former Hi-Lands Shopping Center which will have 686 mixed-income housing units and 38,000 square-feet of leasable commercial space.

The neighborhood is also adjacent to two grocery stores and local leaders are pushing for increased public transit projects and access to the neighborhood.
A podcast for King County residents featuring top local stories.
Applicants must be between 18-25 years of age at the time of appointment.
A truck had damaged the bridge in June 2022 when it collided with it.

source

The Data Science Lab
Machine learning with deep neural techniques has advanced quickly, so Dr. James McCaffrey of Microsoft Research updates regression techniques and best practices guidance based on experience over the past two years.
A regression problem is one where the goal is to predict a single numeric value. For example, you might want to predict the annual income of a person based on their sex, age, state where they live and political leaning (conservative, moderate, liberal).
A previous article series in Visual Studio Magazine have explained regression using PyTorch. But machine learning with deep neural techniques has advanced quickly. This article updates regression techniques and best practices based on experience over the past two years.
A good way to see where this article is headed is to take a look at the screenshot of a demo program in Figure 1. The demo begins by loading a 200-item file of training data and a 40-item set of test data. Each tab-delimited line represents a person. The fields are sex (male = -1, female = +1), age, state of residence, annual income and politics type. The goal is to predict income from sex, age, state, and political leaning.
After the training data is loaded into memory, the demo creates an 8-(10-10)-1 neural network. This means there are eight input nodes, two hidden neural layers with 10 nodes each, and one output node.
The demo prepares to train the network by setting a batch size of 10, Adam (adaptive momentum) optimization with a learning rate of 0.01, and maximum training epochs of 1,000 passes through the training data. The meaning of these values and how they are determined will be explained shortly.
The demo program monitors training by computing and displaying loss values. The loss values slowly decrease which indicates that training is probably succeeding. The magnitude of the loss values isn’t directly interpretable; the important thing is that the loss decreases.
After 1,000 training epochs, the demo program computes the accuracy of the trained model on the training data as 91 percent (182 out of 200 correct). The model accuracy on the test data is 85 percent (34 out of 40 correct). For regression problem accuracy, you must specify how close a prediction must be to the true value in order to be counted as a correct prediction. For the demo, a predicted income that’s within 10 percent of the true value is counted as a correct prediction.
After evaluating the trained network, the demo predicts the income for a person who is male, 34 years old, from Oklahoma, who is a political moderate. The prediction is $45,392.60.
The demo concludes by saving the trained model to file so that it can be used later without having to retrain the network from scratch. There are two main ways to save a PyTorch model. The demo uses the save-state approach.
This article assumes you have a basic familiarity with Python and intermediate or better experience with a C-family language but does not assume you know much about PyTorch or neural networks. The complete demo program source code and data can be found here.
Installing PyTorch
The demo program was developed on a Windows 10/11 machine using the Anaconda 2020.02 64-bit distribution (which contains Python 3.7.6) and PyTorch version 1.12.1 for CPU. Installing PyTorch is like swimming — easy once you know how but difficult if you haven’t done it before.
I work at a large tech company and one of my job responsibilities is to deliver training classes to software engineers and data scientists. By far the biggest hurdle for people who are new to PyTorch is installation.
There are dozens of different ways to install PyTorch on Windows. The configuration I strongly recommend for beginners is to use the Anaconda distribution of Python and install PyTorch using the pip package manager. The Anaconda distribution of Python contains a base Python engine plus over 500 add-in packages that have been tested to be compatible with one another.
After you have a Python distribution installed, you can install PyTorch in several different ways. I recommend using the pip utility, which is installed as part of Anaconda. Briefly, you download a .whl (“wheel”) file to your local machine, open a command shell and issue the command “pip install (whl-file-name).”
You can find detailed step-by-step instructions for installing Anaconda Python for Windows 10/11 here. You can find detailed instructions for downloading and installing PyTorch 1.12.1 for Python 3.7.6 on a Windows CPU machine here.
Preparing the Data
The raw demo data looks like:
There are 240 lines of data. Each line represents a person. The five fields are sex (M, F), age, state of residence (Michigan, Nebraska, Oklahoma), annual income and politics type (conservative, moderate, liberal). The data is artificial. The raw data was split into a 200-item set for training and a 40-item set for testing.
The raw data must be encoded and normalized. The result is:
The variable to predict is income, which is a numeric value rather than a categorical value such as sex or political leaning. Notice that the income values are normalized by dividing the raw values by $100,000. In theory this isn’t necessary, but normalizing regression target values to a -1.0 to +1.0 range or -10.0 to +10.0 range usually helps.
Because neural networks only understand numbers, the sex, state and political leaning predictor values (often called features in neural network terminology) must be encoded. The sex values are minus-one-plus-one encoded rather than 0-1 encoding. In theory both encoding schemes work for binary predictor variables but in practice minus-one-plus-one encoding often produces a better model. For an explanation, go here.
The demo normalizes the age values by dividing each raw value by 100. The technique of normalizing numeric predictors by dividing them by a constant does not have a standard name. Two other normalization techniques are called min-max normalization and z-score normalization. I recommend using the divide-by-constant technique whenever possible. There is convincing (but currently unpublished) research that indicates divide-by-constant normalization usually gives better results than min-max normalization or z-score normalization. The topic is quite complex. For details, go here.
The state of residence values are one-hot encoded as Michigan = (1 0 0), Nebraska = (0 1 0) and Oklahoma = (0 0 1). The order of the encoding is arbitrary. If the state variable had four possible values, then the encodings would be (1 0 0 0), (0 1 0 0) and so on. The political leaning values are one-hot encoded as conservative = (1 0 0), moderate = (0 1 0) and liberal = (0 0 1).
The demo preprocesses the raw data by normalizing numeric values and encoding categorical values. It is possible to normalize and encode training and test data on the fly, but preprocessing is usually a simpler approach.
Overall Program Structure
The overall structure of the demo program is presented in Listing 1. The demo program is named people_income.py. The program imports the NumPy (numerical Python) library and assigns it an alias of np. The program imports PyTorch and assigns it an alias of T. Most PyTorch programs do not use the T alias but my work colleagues and I often do so to save space. The demo program indents using two spaces rather than the more common four spaces, again to save space.
Listing 1: Overall Program Structure
The global device is set to “cpu.” If you are working with a machine that has a GPU processor, the device string is “cuda.” Most of my colleagues and I develop neural networks on a local CPU machine, then if necessary (huge amount of training data or huge neural network), push the program to a GPU machine and train it there.
The demo has a program-defined PeopleDataset class that stores training and test data. Data in a Dataset object can be served up in batches for training by using the built-in DataLoader object. It is possible to use training and test data directly instead of using a Dataset, but such problem scenarios are rare and I recommend using a Dataset for most problems.
The regression neural network is implemented in a program-defined Net class. The Net class inherits from the built-in torch.nn.Module class, which supplies most of the neural network functionality. Instead of using a class to define a PyTorch neural network, it is possible to create a neural network directly using the torch.nn.Sequential class. Using Sequential is simpler but less flexible than using a program-defined class. The fact that there are two completely different ways to define a PyTorch neural network can be confusing for beginners.
In a neural network regression problem, you must implement a program-defined function to compute the accuracy of the trained model. The demo program defines an accuracy() function that works line-by-line, and an accuracy_x() function that works on all data at once.
The demo implements a program-defined train() function. In most cases, I just place all training code directly inside the main() function, but using a train() function is preferred by many of my colleagues.
All of the demo program control logic is contained in a program-defined main() function. The demo program begins by setting the seed values for the NumPy random number generator and the PyTorch generator. Setting seed values is helpful so that demo runs are mostly reproducible. However, when working with complex neural networks such as Transformer networks, exact reproducibility cannot always be guaranteed because of separate threads of execution.
The Dataset Definition
The demo Dataset definition is presented in Listing 2. A Dataset inherits from the torch.utils.data.Dataset class and you must implement three methods: __init__(), __len__(), and __getitem__(). The __init__() method loads the data from file into memory as PyTorch tensors. The __len__() method tells the DataLoader object that uses the Dataset how many items there so that the DataLoader knows when all items have been processed during training. The __getitem__() method returns a single data item, rather than a batch of items as you might have expected.
Listing 2: Dataset Definition
Defining a PyTorch Dataset is not trivial. You must define a custom Dataset for each problem/data scenario. The __init__() method accepts a src_file parameter that tells the Dataset where the file of training data is located. The predictor values and the target income values are read into tmp_x and tmp_y as NumPy matrices in two separate passes. Then the NumPy matrices are converted to PyTorch tensors.
An alternative design is to read predictor and target values in one pass, then extract. Code would look like:
The demo reads data using the NumPy loadtxt() function. Commonly used alternatives include the NumPy genfromtxt() function and the Pandas read_csv() function.
The call to loadtxt() specifies argument comments=”#” to indicate that lines beginning with “#” are comments and should be ignored. The “#” character is the default for comments and so the argument could have been omitted.

Printable Format
For search, devs can select folders to include or exclude.
“If you are doing #Blazor Wasm projects that are NOT aspnet-hosted, how are you hosting them? Would this be useful for you — comment on the issue and what you might expect in the containerization of a Blazor Wasm project?
Machine learning with deep neural techniques has advanced quickly, so Dr. James McCaffrey of Microsoft Research updates regression techniques and best practices guidance based on experience over the past two years.
As the GitHub Copilot “AI pair programmer” shakes up the software development space, Microsoft’s Mads Kristensen reminds folks that Visual Studio’s IntelliCode ain’t too shabby, either.
Microsoft is offering new Visual Studio VM images on its Azure cloud computing platform, some supporting the Dev Box service for cloud-based workstations customized for software development.
> More Webcasts
Problems? Questions? Feedback? E-mail us.

source

MusicRadar is supported by its audience. When you purchase through links on our site, we may earn an affiliate commission. Here’s why you can trust us.
25 mini keys are just the tip of the control iceberg
Donner is upping its portable MIDI keyboard game with the release of the DMK-25 Pro, a feature-rich 25-note controller.
The all-black keys are mini ones that respond to velocity, as do the eight backlit pads. Further control options include ptchbend and modulation touchstrips, four assignable knobs, four assignable sliders, eight function buttons, six transport buttons and an OLED display.
Dynamic Scale Mode, meanwhile, locks the keyboard to specific notes and scales so your playing won’t sound ‘wrong’. This gives you access to 16 modal scales that Donner says are suitable for playing Chinese, Japanese and Middle Eastern music.
Other performance features fall under the Real-time Multi-Mode Control banner – an arpeggiator, note repeat and a chord generation function.
The DMK-25 draws its power over a USB-C port, and also offers a 3.5mm jack MIDI out for connection to mini hardware synths and the like. Its software bundle includes Cubase LE, WaveLab LE and Cubasis LE for iOS, and you also get 40 free keyboard lessons and one beginner course courtesy of online music tuition platform Melodics.
Find out more on the Donner (opens in new tab) website. The DMK-25 Pro costs $110/£125. 
I’m the Group Content Manager for MusicRadar, specialising in all things tech. I previously spent eight years working on our sister magazine, Computer Music. I’ve been playing the piano, gigging in bands and failing to finish tracks at home for more than 30 years, 20 of which I’ve also spent writing about music technology. 
How to record an electric guitar amp
Audiomodern’s Loopmix promises to remix your sample library
Polymetric powers and TR-style sequencing await in the new LL8 2 Eurorack sequencer from Robaux
Don’t miss the latest deals, news, reviews, features and tutorials
Thank you for signing up to Musicradar. You will receive a verification email shortly.
There was a problem. Please refresh the page and try again.
MusicRadar is part of Future plc, an international media group and leading digital publisher. Visit our corporate site (opens in new tab).
© Future Publishing Limited Quay House, The Ambury, Bath BA1 1UA. All rights reserved. England and Wales company registration number 2008885.

source

28 Oct 2022 12:22 PM GMT
The audio which is 27 minutes long, has voices of Nanda Kumar, Rama Chandra Bharathi alias Satish Kumar and the seer Simhayajulu (Photo: Tv9 Telugu)
A phone conversation between the three accused in TRS MLAs poaching case has been released and accessed by TV9 Telugu. The audio, which is 27 minutes long, has voices of Nanda Kumar, Rama Chandra Bharathi alias Satish Kumar and the seer Simhayajulu.
In the audio recording, a discussion can be heard about the money that is to be given to the four MLAs. Referring to MLA Pilot Rohit Reddy, Bharathi says ”He is expecting Rs 100 cr. Others who may come with him may demand less. Just nominally.”
The seer can be heard saying that the operation was ‘bypassed by the state’ and that the operation was being done by ‘Delhi’. Rama Chandra Bharathi says that BJP leaders Bandi Sanjay and Kishan Reddy are unlikely to know about the operation.
Earlier on Friday, a recording of a conversation between the seer and MLA pilot Rohith Reddy was released. In that conversation, the two are heard exchanging information about their meeting.
On Wednesday, the Cyberabad Police raided the Moinabad farm house and arrested the three accused for allegedly trying to poach four TRS MLAs.
The local ACB court on Thursday said that the police cannot keep the accused in remand as there was no cash seized from the farm house. BJP leaders denied that they have any involvement in the issue and said that they want a CBI investigation or a judicial investigation.
Suryakumar reaffirmed his status as the world’s number one T20 batter with another magnificent effort as India crushed…
After Suryakumar Yadav’s blazing half-century powered the Asian giants to a massive total of 186/5, Indian bowlers…
Over 200 employees mostly from the engineering, marketing and communications department were sacked in a day without…
TRS candidate Kusukuntla Prabhakar Reddy has cumulatively polled 96,598 votes after fifteen rounds.
Modi said the forces that indulged in spreading hatred and defamed Gujarat will be swept out of the state in the next…
Addressing back-to-back election rallies in the state which votes to elect another government on November 12, Shah said…
India finished on top of group 2 and will now play their fourth T20 World Cup semifinal against England in Adelaide on…
Standing on the verge of an early exit from the tournament after defeats against India and Zimbabwe in the Super 12…
England’s nervy win knocked defending champions and hosts Australia out of the 2022 T20 World Cup.
The screening of Janhvi Kapoor and Sunny Kaushal-starrer happened in Mumbai on November 3.
RSSB, also known as Dera Baba Jaimal Singh, is located in Beas town, nearly 45 km away from Amritsar city, and has…
The movie has been by filmmaker Gurmmeet Singh.

source

Paramore have shared plans for an expansive 2023 North American arena tour. They’ll embark on the 27-date run on May 23 at Charlotte, North Carolina’s Spectrum Center and close it out on August 2 at St. Paul, Minnesota’s Xcel Energy Center, snaking across the U.S. (and Toronto) along the way. They’ll have alternating support on their cross-country trek from Bloc Party, Genesis Owusu, Foals, and The Linda Lindas. Fans can by tickets early via two unique presales — one this coming Monday (November 7) at 11:59 p.m. Eastern Time, the other on Thursday, November 10 at 7:30 a.m. local time — with general sales starting Friday, November 11 at 10 a.m. local time.
The tour will follow the February 10, 2023 release of This Is Why, Paramore’s sixth studio album and their first since 2017’s After Laughter. We added the record’s lead single and title track to our Songs You Need In Your Life playlist last month.
The band is currently finishing another tour through North America (announced back in July), after which they’ll head to South America and the United Kingdom, with a show each in Glendale, Arizona and Dublin, Ireland in between. Check out their 2023 arena tour poster and view all of their upcoming live dates below.
November 7 – Toronto, ON – HISTORY
November 9 – Chicago, IL – The Chicago Theatre
November 11 – Cincinnati, OH – The Andrew J Brady Music Center
November 15 – Atlanta, GA – Tabernacle
November 16 – St. Augustine, FL – The St. Augustine Amphitheatre
March 2 – Santiago De Surco, PE – Arena Perú
March 5 – Santiago, CL – Movistar Arena
March 7 – Buenos Aires, AR – Movistar Arena
March 9 – Rio De Janeiro, BR – Qualistage
March 11 – São Paulo, BR – Centro Esportivo Tiete
March 14 – Bogotá, CO – Movistar Arena
March 18 – Glendale, AZ – University of Phoenix Stadium
April 13 – Dublin, IE – 3Arena
April 15 – Cardiff, WLS – Cardiff International Arena
April 17 – Glasgow, SCT – OVO Hydro
April 18 – Manchester, UK – AO Arena
April 20 – London, UK – The O2
April 22 – Birmingham, UK – Utilita Arena
April 23 – London, UK – The O2
May 23 – Charlotte, NC – Spectrum Center*×
May 25 – Atlanta, GA – State Farm Arena*×
May 27 – Atlantic City, NJ —Adjacent Festival!
May 30 – New York, NY – Madison Square Garden*×
June 2 – Washington, DC – Capital One Arena*×
June 4 – Cleveland, OH – Rocket Mortgage Fieldhouse*×
June 5 – Indianapolis, IN – Gainbridge Fieldhouse*×
June 7 – Detroit, MI – Little Caesars Arena*×
June 8 – Toronto, ON – Scotiabank Arena*×
June 10 – Columbus, OH – Schottenstein Center*×
June 11 – Pittsburgh, PA – PPG Paint Arena*×
June 13 – Orlando, FL – Amway Center*×
June 14 – Hollywood, FL – Hard Rock Live*×
July 6 – New Orleans, LA – Smoothie King Center+°
July 8 – Fort Worth, TX – Dickies Arena+°
July 9 – Austin, TX – Moody Center+°
July 11 – Houston, TX – Toyota Center+°
July 13 – Denver, CO – Ball Arena+°
July 16 – San Diego, CA – Viejas Arena+
July 19 – Los Angeles, CA – Kia Forum+
July 22 – San Francisco, CA – Chase Center+
July 24 – Seattle, WA – Climate Pledge Arena+°
July 25 – Portland, OR – Veterans Memorial Coliseum+°
July 27 – Salt Lake City, UT – Vivint Arena+°
July 29 – Tulsa, OK – BOK Center+°
July 30 – St Louis, MO – Enterprise Center+°
August 2 – St. Paul, MN – Xcel Energy Center+°
* with Bloc Party
× with Genesis Owusu
+ with Foals
° with The Linda Lindas

source

 
 
On Nov. 1, 2022, singer-songwriter Taylor Swift announced her latest tour, which is henceforth known as The Eras Tour. Marking over 15 years since her first-ever album, The Eras tour will transport attendees through several musical eras of the 32-year-old’s illustrious career. Swift’s discography spans multiple genres — ranging from country and folk to synth-pop and country pop — and her lyrics often convey difficult emotions and scenarios, such as heartbreak, loss, grief, and sorrow. After the announcement was made, Swift’s social media following immediately went crazy for the opportunity to see the superstar live — whether it be the first time or the fifth time.
Swift hasn’t been on tour since 2018 when she performed songs from her album Reputation. Sadly, she was supposed to tour in 2020, but unforeseen circumstances resulting from the COVID-19 pandemic meant that her stadium performances to promote Lover were canceled. Following the immense success of her studio albums Folklore and Evermore, Swift — during her hiatus from touring — has released three more studio albums in Midnight and deluxe re-recordings of Fearless and Red.
The Eras Tour will begin in March 2023 in the US, beginning in Glendale, AZ. Swift has even snagged the likes of Paramore, Phoebe Bridgers, GAYLE, and HAIM to open.
On Nov. 4, just three days later, Swift posted an updated tour date sheet, which added eight more dates to the schedule.
UM. Looks like I’ll get to see more of your beautiful faces than previously expected… we’re adding 8 shows to the tour 😆https://t.co/KFuqvr0hhO pic.twitter.com/4LTYSnwKJO
All dates highlighted in blue were recently added, all of which were extra nights in the same locations, likely in preparation for tickets to sell out on the other days. The dates include the likes of Nashville, TN; Tampa, FL; Philadelphia, PA; Foxborough, MA; East Rutherford, NJ; Seattle, WA; Santa Clara, CA; and Los Angeles, CA.
Here is the complete list of tour dates — including the recent additions — and their opening acts:

03/18 – Glendale, AZ @ State Farm Stadium (with Paramore, GAYLE)
03/25 – Las Vegas, NV @ Allegiant Stadium (with beabadoobee, GAYLE)
04/01 – Arlington, TX @ AT&T Stadium (with beabadoobee, Gracie Abrams)
04/02 – Arlington, TX @ AT&T Stadium (with beabadoobee, Gracie Abrams)
04/14 – Tampa, FL @ Raymond James Stadium (with beabadoobee, Gracie Abrams)
04/15 – Tampa, FL @ Raymond James Stadium (with beabadoobee, Gracie Abrams)
04/22 – Houston, TX @ NRG Stadium (with beabadoobee, Gracie Abrams)
04/28 – Atlanta, GA @ Mercedes-Benz Stadium (with beabadoobee, Gracie Abrams)
04/29 – Atlanta, GA @ Mercedes-Benz Stadium (with beabadoobee, Gracie Abrams)
05/05 – Nashville, TN @ Nissan Stadium (with Phoebe Bridgers, GAYLE)
05/06 – Nashville, TN @ Nissan Stadium (with Phoebe Bridgers, GAYLE)
05/12 – Philadelphia, PA @ Lincoln Financial Field (with Phoebe Bridgers, GAYLE)
05/13 – Philadelphia, PA @ Lincoln Financial Field (with Phoebe Bridgers, GAYLE)
05/14 – Philadelphia, PA @ Lincoln Financial Field (with Phoebe Bridgers, GAYLE)
05/19 – Foxborough, MA @ Gillette Stadium (with Phoebe Bridgers, GAYLE)
05/20 – Foxborough, MA @ Gillette Stadium (with Phoebe Bridgers, GAYLE)
05/21 – Foxborough, MA @ Gillette Stadium (with Phoebe Bridgers, GAYLE)
05/26 – East Rutherford, NJ @ MetLife Stadium (with Phoebe Bridgers, GAYLE)
05/27 – East Rutherford, NJ @ MetLife Stadium (with Phoebe Bridgers, Gracie Abrams)
05/28 – East Rutherford, NJ @ MetLife Stadium (with Phoebe Bridgers, Gracie Abrams)
06/02 – Chicago, IL @ Soldier Field (with girl in red, OWENN)
06/03 – Chicago, IL @ Soldier Field (with girl in red, OWENN)
06/10 – Detroit, MI @ Ford Field (with girl in red, OWENN)
06/17 – Pittsburgh, PA @ Acrisure Stadium (with girl in red, OWENN)
06/24 – Minneapolis, MN @ U.S. Bank Stadium (with girl in red, OWENN)
07/01 – Cincinnati, OH @ Paycor Stadium (with MUNA, Gracie Abrams)
07/08 – Kansas City, MO @ GEHA Field at Arrowhead Stadium (with MUNA, Gracie Abrams)
07/15 – Denver, CO @ Empower Field at Mile High (with MUNA, Gracie Abrams)
07/22 – Seattle, WA @ Lumen Field (with HAIM, Gracie Abrams)
07/23 – Seattle, WA @ Lumen Field (with HAIM, Gracie Abrams)
07/28 – Santa Clara, CA @ Levi’s® Stadium (with HAIM, Gracie Abrams)
07/29 – Santa Clara, CA @ Levi’s® Stadium (with HAIM, Gracie Abrams)
08/03 – Los Angeles, CA @ SoFi Stadium (with HAIM, OWENN)
08/04 – Los Angeles, CA @ SoFi Stadium (with HAIM, OWENN)
08/05 – Los Angeles, CA @ SoFi Stadium (with HAIM, GAYLE)
Tickets go on sale on Nov. 15, but Swifties have two options for pre-sale registration:  and .

source

Nici Forner and Anna Schlaff pose for a photo in front of Studio One Thirty Four at 134 W. Middle St. Suite A.
CHELSEA, MI — Studio One Thirty Four is offering the Chelsea community not only a full service hair salon, but also a “comfortable environment” for its clients and stylists, said owner and stylist Nici Forner.
Studio One Thirty Four began at 134 W. Middle St. Suite A in March, but the business plans to celebrate its recent opening with a ribbon-cutting ceremony at 5 p.m. Wednesday, Nov. 9. The event is a chance to meet the studio’s stylists and learn more about what the salon is all about.
“I always worked at salons in Chelsea, but I just wanted to do something that was different for me,” Forner said. “I just wanted to have my own small studio space where there’s room for growth (and is) something different than what I’ve done before.”
Forner, who has around 18 years of experience, works alongside another stylist, Anna Schlaff, to bring services like women’s haircuts, men’s haircuts, custom coloring, highlighting, styling and more. Forner added that custom coloring and creating bridal hairstyles are the studio’s “strong suit.”
The pair of stylists work with clients to meet their goals and expectations, Forner said, adding that she has plans to continue to add more stylists to Studio One Thirty Four in the future.
Although Forner could have opened her salon in any other town, she decided she wanted to continue her efforts in the city she calls home.
“I love Chelsea because I feel like wherever you go you can see a familiar face,” she said. “Everything’s very familiar and comfortable which is something that I want to continue to perpetuate even with a new salon.”
Forner said the studio’s hours vary depending on the day from Monday to Saturday, and that early morning and late evening bookings can be scheduled by appointment. Currently, the studio is taking walk-in appointments from 10 a.m. to 6 p.m. Tuesdays and Thursdays and 10 a.m. to 3 p.m. Saturdays.
To learn more about Studio One Thirty Four, visit its website or its social media.
Read more from The Ann Arbor News:
Former Fingerle Lumber Company reimagined as Union Rec restaurant
Write-in candidate in Ann Arbor mayoral race says city has gotten too corporate
Your guide to what’s on the Nov. 8 general election ballot in Washtenaw County
If you purchase a product or register for an account through one of the links on our site, we may receive compensation.
Registration on or use of this site constitutes acceptance of our User Agreement, Privacy Policy and Cookie Statement, and Your California Privacy Rights (User Agreement updated 1/1/21. Privacy Policy and Cookie Statement updated 7/1/2022).
Cookie Settings
© 2022 Advance Local Media LLC. All rights reserved (About Us).
The material on this site may not be reproduced, distributed, transmitted, cached or otherwise used, except with the prior written permission of Advance Local.
Community Rules apply to all content you upload or otherwise submit to this site.
Ad Choices

source

MusicRadar is supported by its audience. When you purchase through links on our site, we may earn an affiliate commission. Here’s why you can trust us.
The electronic musician, label head and plugin developer on the self-built tools behind his new project, Paperclip Minimiser
Electronic musician, label head and developer John Howes’ music lives on the boundary between chaos and control. 
On the one hand, his work is largely produced using systems and tools of his own design: rhythms and notes are generated using Strokes, a sequencing environment he’s developed, synth patches are meticulously constructed on his Nord Modular G2 and Elektron Machinedrum, while parameters are modulated using another self-built software tool, the global modulation matrix Dispatch. If he discovers a plugin or an instrument that sounds intriguing, unlike many of us, his inclination isn’t to buy it – it’s to build his own version, often creating something new and unexpected in the process.
Though he’s rigorous in his command of the creative process, Howes tells us, it’s when he begins to let go that the magic happens. Captivated by the limitless potential of generative music-making, he carefully integrates elements of chance and indeterminacy into these highly structured systems, exploring the musical possibilities presented by implanting ghosts in his machines – cybernetic, semi-autonomous agents that possess what he’s called “lo-fi AI”. 
His latest release, a self-titled project under a new alias, derives its name from Paperclip Maximiser, a thought experiment that envisions how an AI tasked solely with manufacturing paperclips could run out of control and inadvertently kill us all in the process, were it not instilled with some form of self-correcting, machine-based ethics. 
Thankfully, the AI-powered tools used to produce Howes’ latest release, Paperclip Minimiser, haven’t yet threatened us with extinction: instead, they’ve helped him record some of the best experimental music we’ve heard this year. Across eight untitled tracks, the project explores the outer reaches and inner depths of electronic esoterica, venturing through dubwise techno abstraction, Hassell-esque fourth-world sonics and corrosive machine jams of the highest order. 
We caught up with John Howes to hear more about the ideas behind his new record, the music software he’s developed as Cong Burn, and his fascination with the Nord Modular. 
How did you get into music-making initially?
“I started making tunes on PlayStation 1 and cracked software, FL Studio on PC. When I moved to Manchester when I was 18 I got into a bit more and started buying equipment – drum machines and effects and modular stuff, and now I’m kind of in between all of it. I’m still using all these things – not so much cracked software, though [laughs]. But still torn between the world of drum machines, synths, hardware, and software.”
When did you set up Cong Burn and what were you setting out to do with the label?
“Basically, all of my friends make tunes but none of them were putting any of it out. I’d had a couple records out, and I knew the good and bad sides, and the things that my friends would and wouldn’t want to do in the music industry. So as a safe way of getting their music out, I started releasing their stuff. Originally it was meant to be a radio show, where we used to rent the studio and we’d get together and record stuff. It was supposed to be a monthly radio show, but we did it once and it took three months. [laughs] So we ended up doing tapes, then started doing 12”s and parties and stuff. It all grew out of like three or four friends from the Northeast, where I’m from, then it just grew and grew, and people have come and gone along the way.”
When did you start getting into development?
“When I was at uni I made a bunch of Max for Live devices. I had one of those Eurorack converter things where you can send CV out. I didn’t have a lot of money, as a student, so I was making LFOs and sequencers in Max that would output CV into my modular, basically as a way of saving money. And then I didn’t touch it for five or six years.
Pretty much all the music that I’ve made in the past three or four years has started with Strokes in some capacity
“I worked in the music industry for five years, and I got to the stage where I was like, the next job that I get in this industry is going to be working on stuff that I don’t give a shit about. Mainstream, money-making, business techno or pop music type of thing. So it was like, I guess I’m at the end of my road here. I just left the music industry completely, didn’t put out a record or do any work on music for about a year, and studied coding. Then, the first job that I got in coding was for Behringer. So I kind of went straight back in.”
Where did the idea come from for Strokes, the new plugin you’ve developed?
“I was reading a book about Jaki Liebezeit, the drummer from Can. He’s an absolutely sick drummer, and he’s got this really repetitive style, with really complex rhythms. I wanted to make music that sounded like that, but I didn’t want to program it in MIDI clips in Live, it’d take fucking ages. So I started looking at these ways of devising rhythms, and came across Euclidean rhythms. 
“When you look at the stuff Jaki was doing, he never says the word Euclidean, but it’s all there – all of it is about mathematical relationships, and the diagrams he’s drawn are Euclidean rhythms, he just didn’t call them that. So I started working on it mainly as a way of generating interesting drum patterns that constantly evolve. 
“Before I started working on it, I used to be into Eurorack stuff, I had this big Eurorack system made of all Doepfer stuff. I used to build mini versions of Strokes on that. I made a hardware version, years and years ago, and had this whole system patched for a year, and was constantly generating stuff with it. Then I turned it into a Max thing, later on, and it’s been in development for nearly four years now. It started as a small thing to generate rhythms, and it’s grown into a full-on sequencing environment. Pretty much all the music that I’ve made in the past three or four years has started with it in some capacity.”
Tell us about the other plugin you’ve worked on, Dispatch.
“Ableton doesn’t have a proper patch bay, but Bitwig has a global modulation system. You can stack modulators on top of each other, so you can have an LFO and a sequencer both on the same dial. Ableton doesn’t let you do that stuff, so Dispatch basically started out as a matrix mixer, where you can hook up four live LFOs into a matrix mixer, and then mix those LFOs into four different destinations. 
“That was version one, and I used that for ages. I sent it to a couple of people who were like, this is cool, but I don’t want to make four LFOs every time I use it. So I started to think then about how I would design my own LFOs, and what would I want those LFOs to look like? Serge stuff was a big influence on the modulators. The Dual Universal Slope Generator is like the rise and fall generator, basically. You can do exponential versus linear curves and stuff like that, and that’s all taken from Serge modular stuff. Cong Burn’s Strokes MIDI sequencing plugin is an instant inspiration generator
“It started out as a mixer, and then I ended up adding modulators to it. From there it grew to having MIDI input and output, audio input and output, and became like a MIDI-to-CV converter. I kind of use it on everything, but only a little bit. It’s a useful tool to have. If you’re modulating a filter in Live, and you want to modulate a hardware filter, you can hook it up to Dispatch and get the exact same CV signal sent out, and patch it in. It just seemed like a big gap in Ableton that they didn’t have this thing figured out yet.”
Is Bitwig your DAW of choice, then?
“No, I’m still using Ableton actually. I use it in quite a limited capacity. I don’t use any of the built-in synths, but I use a lot of the built-in effects. The more I learn about old bits of hardware and stuff – I’ve been researching old Prophet samplers, like the 2002, which I’m pretty sure Monolake had back in the day. Then when you learn about that, and then you look at Live’s Sampler, it’s like, oh man, these are exactly the same! 
“The Live Sampler has all the features of this Prophet 2002 which Robert Henke had back in the day. Things like zero-crossing the sample points and loops, things like that. Ableton’s cool because it’s kind of like having Robert Henke’s old studio in software. [laughs] So I mostly use Ableton as a recorder, and a modulation matrix type of thing. I generate most of my sounds using hardware.”
Tell us about the new record. What’s behind the decision to release under a new alias?
“It’s quite different to a lot of the previous things that I’ve done, in terms of process. The reason that I put it under this new name is that, at the core of all the tracks on this new record, I’m into the idea of trying to put semi-artificial intelligence into the generative system. 
I was trying to give each voice or each section of a track its own space, almost like it’s an actor in a play. Almost like Curb Your Enthusiasm ambient music
“Most of the music that I make comes from complex generative systems made using Strokes, Dispatch and other bits of equipment. But in the techniques that I use on this record, the focus was on trying to give each voice or each section of a track its own space, almost like it’s an actor in a play, or something like that. Almost like Curb Your Enthusiasm ambient music, where everything knows where it’s going, but as soon as someone actually hits play, you don’t know exactly what the results are gonna be. 
“The idea is that all the instruments, all the voices and all the parts all listen to each other and respond to what each other are doing. But also they have to have their own intelligence in a way, they make their own decisions. That’s where the name comes from. It’s called Paperclip Minimizer. Paperclip Maximizer is this thought experiment about an AI that is given the job of making paperclips and destroys the universe. So this is a self-deprecating name, where it’s struggling to make ambient music. [laughs]”
“The core of it is, I try to build these systems where each voice or each actor has its own decision-making abilities. You can see this in Strokes. The Shares module in Strokes has this stuff built into it, but it’s a simple lofi version of an AI Markov chain, where the input is constantly feeding into the next step. It counts the note triggers on every single channel, then you can set a designation of how much you want on each channel, and over time, that designation will change. So if you set all the faders to 50% and hit play, that’s when you get this feeling like there’s a ghost in the machine. You’ll have loads of snares for a second, and then that will trigger the snares to stop and the hi-hats to start… once the system’s moving, that’s when I find the most interesting results.”
What is it about using generative methods you find so stimulating?
“One of the things that I find really special about it is that I don’t know the outcomes before they occur. So when I’m listening to things, and I’m recording these jams, I’m experiencing them for the first time. In my older work, I would just record these flashes of like: this is starting to sound good, so I’ll hit record. Then the album would be like a collection of these six-minute long recordings of longer jams where everything seemed to land right. 
“But it’s kind of moved on from there now, where the reason that I’m using generative systems now is that I’m a control freak and I want to know everything that’s going on. I’m not a very good collaborator, so I’ve kind of built these systems where I have five collaborators in a track for different instruments, or different voices that are all connected, and I can kind of jam with the machines instead of having to jam with people. [laughs]”
As if the machines are improvising with you?
“Exactly. Give them a little bit of room where they can do their own thing, but it’s all part of a really structured system. I used to have a modular system that I performed live with for two or three years. It had three different voices going into it, was 84 HP, it was dead small. But I knew the synth so well that I could hear the sound in my head just by looking at it. It’s kind of similar with Strokes and the systems I have going on now. I can tell you 90% how it’ll sound without hitting play, but it’s that final 10%…”
You’ve described the set-up you used for this release as an ‘authentic 2006 studio.’ Were you intentionally trying to place yourself in a different era?
“Not consciously, but it is true that a lot of the music that I was listening to during that period was from the early 2000s. Süd Electronic, Mille Plateaux, Raster-Norton and Source Records, these sorts of labels. It’s more of a funny coincidence than it is anything deliberate. I use all sorts of different equipment, although the core of the record was made on a Nord Modular, Machinedrum and Monomachine, I do have some other bits around that sometimes filter into it. And still use the occasional plugin, very occasionally. So yeah – it’s not by design, not intentional, but it’s there.”
You’ve mentioned that you mostly used the Nord Modular G2, what is it about this synth that you found inspiring?
“Oh, man. The Nord lets you design your own drum machines, synthesizers, effects and sequencers. It’s almost like VCV Rack, but the modules haven’t been updated for 15 years. With the Nord, I get the sense that they threw everything they had into it. All the best engineers were working on it, and they basically went bust trying to make it. The amount of engineering expertise that went into this machine made it one of the best machines out there. All of the oscillators, all the tools, everything sounds great. 
A new synth will come out, but rather than having to spend any money and buy it, I can try to build a close approximation on the Nord
“There’s just enough in there that you could spend your entire life digging into these things. Now that I’ve got it, I kind of don’t want to buy any more equipment ever again – every synth that I want, I can just try and make it on the Nord. So I’ve got a Lyra-8 that I made in there, and I’m gonna make a Bastl softPop tomorrow. A lot of the time a new synth’ll come out, but rather than having to spend any money and buy it, I can try to build a close approximation on the Nord and usually end up somewhere else that I didn’t expect.
“It’s not an easy machine to use. It’s hard to describe, but it feels like you go to work every time you use it. When I bought it, I knew this machine was gonna take over, so I basically booked like two months off, and Monday to Friday, nine to five, I was working on the Nord. The reason that I got it in the first place is that all the patches are shared online, so I was going to dig through some online stuff and use it as a prototyping machine. Dig through the history of the Nord and all the things people built with it and see if there’s any stuff in there that I can take inspiration from for the next Strokes, or the next Dispatch. 
“So I bought it as a prototyping environment, and that’s the reason why I spent two months learning it. And I still know 10% of what that machine is capable of. It’s insane. I wouldn’t recommend that everybody pick one up because they’re really expensive and really temperamental, but yeah – holy shit, man, this machine. It’s the end game at the minute.”You also mentioned both the Machinedrum and Monomachine. Are you a big Elektron fan?
“The Monomachine and the Machinedrum, I didn’t use a ton. I didn’t use any of the internal sequencers on either of them. A lot of what I was doing was building sequences on the Nord and sending that out to the Machinedrum and Monomachine. The Machinedrum, I’ve had a couple of them over the years, and they’re amazing. 
“It’s one of those things where I get one and I’ll use it for a year, get sick of the sound of it, and sell it or put it on the shelf. I’m not as interested in the new stuff, but the Monomachine, and the Machinedrum especially – it’s basically a modular synth, you can do a ton of internal routing stuff. Tracks that control other tracks, tracks that are follower tracks to other tracks that’ll take trigger information. 
“There’s infinite things you can do. It’s similar to the Nord in that it’s an open box and you can do what you want with it. Simple tools that you can patch together in really crazy ways. I’m not as deep into the Monomachine, I’m not as keen on the synth engines. I don’t use either that much anymore, to be honest, since I’ve got the Nord. I’ve got close approximations of all the Machinedrum and Monomachine voices that I’ve built myself inside the Nord. Other things are slowly getting pushed out. Maybe in ten years, it’ll just be 10 Nord Modulars. [laughs]”
We’ve covered hardware, but which software tools do you find the most inspiring?
“I generally keep things really simple. I don’t get too lost in the world of plugins. I have a couple of presets for a couple of effects that I’ve used. On this new record there’s an Eventide plugin that I used that I had one preset for, and it’s on half the tracks on here. But yeah, largely, I don’t use other people’s stuff. I try to use my own things, or I try to use Ableton stock stuff. If there’s a thing out there that I want, a Max thing that I think is cool, I’ll just build it myself, and make my own version of it. 
“I don’t generally look for new equipment anymore. If something new and interesting comes along, I’ll just try and do it myself. With the plugin stuff, I’m kind of fairly comfortable with building other people’s stuff, whatever it is. DSP is something I’ve not got massively into yet, but that’s probably going to be the next thing.”
Which other artists would you say have been the most influential on your work?
“When I’m actually working on music, I generally don’t listen to anything else. One of the differences on this new record is that I basically shut myself off from all music culture – no Boomkat, no Discogs, no social media – I was trying to stay disconnected as much as possible. So when it actually comes to working on music, I’ll shut everything out. 
When it actually comes to working on music, I’ll shut everything out
“I don’t mean for a couple of hours, I mean for two weeks I’ll listen to no music other than my own, and sit and work on my own stuff, and then reinstall Instagram after that’s done and get back to normal life, or whatever. But I’m trying to cut off any kind of external influence that might occur. I still listen to music during those periods, but it’s wildly different music. I’ll listen to All Trades on NTS or something that’s totally unrelated – just so that nothing can filter in.
“In terms of musical artists that have had an influence on me, it’s difficult to say. There’s been so many over the years, like O Yuki Conjugate, and the club night meandyou., which is run by Lyster, Herron and Juniper. I’d say right now, my Cong Burn mates and the extended universe of people that send us music is pretty much all I’m really interested in musically.
“About two years ago, we subconsciously started only playing new music on the Cong Burn show, or like 90% new music. And that kind of led to people sending us more music, and one thing led to another, and now we’re getting sent loads of really, really sick music every month, from people like Rastegah and oh!t. It’s like nothing else that’s out there at the minute, and that’s super inspiring. But likewise, I don’t try to make music for that radio show, or anything like that. I’m trying to put off influence as much as I am taking it in.”
Paperclip Minimiser is out now on Peak Oil. 
Strokes and Dispatch are available to purchase from Cong Burn’s website. 
I’m the Tech Features Editor for MusicRadar, working on everything from artist interviews to tech tutorials. I’ve been writing about (and making) electronic music for over a decade, and when I’m not behind my laptop keyboard, you’ll find me behind a MIDI keyboard or a synthesizer. 
Recording vocals – the fundamental principles
DW has unveiled its DWe acoustic/electronic drum set, here’s everything we know about it so far
Audiomodern’s Loopmix promises to remix your sample library
Don’t miss the latest deals, news, reviews, features and tutorials
Thank you for signing up to Musicradar. You will receive a verification email shortly.
There was a problem. Please refresh the page and try again.
MusicRadar is part of Future plc, an international media group and leading digital publisher. Visit our corporate site (opens in new tab).
© Future Publishing Limited Quay House, The Ambury, Bath BA1 1UA. All rights reserved. England and Wales company registration number 2008885.

source

After a scorcher of a year in 2022 (we’re talking about the lineup, not the weather, obviously), Hinterland is back with a list of amazing indie acts on the bill for 2023, with some up-and-coming names.
“Of the big name headliners, I’m most excited about Orville Peck, who’s making his second Hinterland appearance. I saw him play in a smaller club a few years ago but missed his Hinterland 2021 performance, so this is a music festival monkey I’ll be happy to get off my back,” writes IPR’s Tony Dehner. “As for the midday performers, I’ll be sure to head down to the pit for Joy Oladokun and Tomberlin, both of whom make beautiful music. Some of my favorite sets from last year were in the middle of the day: Sierra Ferrell, Briston Maroney, and Miko Marks were all on my radar before Hinterland, and I came away from the weekend a huge fan of all three. I’m looking forward to being surprised again this year.”
In 2022, the festival expanded to four days but going into 2023 is scaling back down to three. Our Studio One crew was honestly pretty excited to see this. Since there is one stage during the day, three days just feels right for this fest.
“Maggie Rogers’ 2022 record Surrender was a big leap forward from her debut album. I somehow missed her 2019 set at Hinterland, but I won’t let that happen again! She might be the artist I’m most looking forward to seeing,” writes IPR All Access and Studio One Tracks host Mark Simmet.

“The strength of this lineup is that nearly all of these artists are at a point where their careers are still growing. In other words, they haven’t peaked yet, and it’ll be exciting to see what they bring to their performances. Bon Iver have been around the longest, but even this band is currently unpredictable. Their most recent work has been very experimental, and I’m really looking forward to their headlining set on the Friday night of the festival,” Mark said. “Sylvan Esso have some really good songs, and their current album indicates their willingness to experiment these days. There may be some surprises in their 2023 set, and I want to be there for it, regardless. I found Angel Olsen’s latest record to be a bit underwhelming, but I’m a fan of her powerful voice. I can see (and hear) her being great on the Hinterland stage.”

We’re spinning music by Wallows and Faye Webster on All Access and Studio One Tracks. Skeggs and spill tab are new adds to our library that we’ll be excited to see live and dig into.
We’re all about the music discovery that happens each summer on Iowa’s festival lineups. Host Cece Mitchell has seen both Wallows and Houndmouth at other venues and says she can’t wait for their Hinterland sets.
“I saw Wallows at a festival last summer, and they put on an amazing live show,” writes Cece. “I was also really starstruck by frontman Dylan Minnette, who you might recognize from his big roles in 13 Reasons Why and The Dropout. Their 2019 track “Are You Bored Yet?” featuring Clairo is one I sing along to in the car all the time.”
Earlier this year, Studio One had the opportunity to interview Matt Myers, the frontman of Houndmouth.
“He was really kind over the phone, which I was absolutely thrilled with because “Sedona” is one of those my friends and I sing together LOUD. I predict that song will be an amazing closer to their set,” Cece laughed.
“I first heard of Faye Webster when Barack Obama put her song “Kingston” on one of his summer playlists. That year, it made my top ten most-played songs of the year! It will be interesting to see if she plays any of the more orchestral songs off her latest EP, Car Therapy Sessions,” she added.

Hinterland has also paid close attention to making sure that music festival parents have a place to take their littles to for a cool down and do something entertaining beyond watching the music on the mainstage. In years past, the Hinterkids tent has hosted “Meet A Musician” talks with Tre Burt and Nathaniel Rateliff. This year, a special music booking just for the kids is on the lineup, and Des Moines artist Ramona Muse Lambert is planning another original edition of the Hinterkids Activity Book.
Koo Koo Kanga Roo was created in 2008 by two friends competing in a battle of the bands at St. Mary’s University in Winona, MN. The idea for the band “centered around crowd participation, comedy and dance music, stretching the concept of what a “band” could be.”
“Of COURSE I’m heading over to Hinterkids for Koo Koo Kanga Roo,” says Tony. “I mean, come on.”
Once again, there are no Iowa artists on the main stage, but a few musicians from our home state played at the brunch stage last year. We might see that again, and possibly over at the campfire stage as well?

Friday, August 4: Bon Iver, Sylvan Esso, Houndmouth, The Regrettes, Skegss
Saturday, August 5: Zach Bryan, Orville Peck, Angel Olsen, Noah Kahan, The Teskey Brothers, Joy Oladokun, Wilderado, Patrick Droney, Riley Downing, Pony Bradshaw, Koo Koo Kanga Roo (Hinterkids Stage)
Sunday, August 6: Maggie Rogers, Wallows, Faye Webster, Del Water Gap, Tomberlin, spill tab, Yot Club
Tickets to the fest go on sale Friday, Nov. 5 at 10 a.m. A presale will begin on Thursday, November 3 at 10am CDT. Fans can sign up for the presale at hinterlandiowa.com/presale.

source