Deep Longevity Building An Arsenal of Aging Clocks
By Deborah Borfitz
January 21, 2021 | In the not-too-distant future, people could be paying a monthly fee to access a long list of “aging clocks” serving as bellwethers of their health and efforts to improve their physical condition. Physicians might also be procuring medical-grade reports on the aging process of their patients. It is a business model Hong Kong startup Deep Longevity is banking on.
Deep Longevity, a spinoff of artificial intelligence (AI) company Insilico Medicine, launched its “longevity as a service” product suite last summer. Five aging clocks have already been integrated into its longevity web platform Young.AI, says project manager Fedor Galkin. Access, for now, is free of charge.
Clocks on the menu currently are BloodAge (using clinical blood tests), PhotoAge (using facial photos), Anamnesis Age (based on lifestyle data), and, most recently, PsychoAge (which predicts chronological age) and SubjAge (which describes personal aging rate perception). “Among these, BloodAge is our ‘killer clock’ since it is easily interpreted by medical professionals and every clinic already has the infrastructure to obtain its input,” says Galkin.
But the PsychoAge and SubjAge clocks are also notable additions because the psychological aspect of aging has been severely understudied. These deep psychomarkers of aging were recently described in a study published in Aging (DOI: 10.18632/aging.202344).
In the study, the two AI-based age predictors were linked to mortality risk. People whose SubjAge was five years greater than their chronological age were twice as likely to die as those with a normal age perception, but SubjAge predictions could be substantially reduced by developing openness to new experiences, keeping the bar high, being productive, and not backing away from difficult-to-reach goals.
The psychological aging clocks were trained on a collection of more than 10,000 questionnaires completed by people who were part of a midlife study by the MacArthur Foundation. The models presented in the publication were reworked into 15-question surveys now available at Young.AI.
Transcriptomic, microbiomic, and DNA methylation aging clocks have also been developed by Deep Longevity but are “a bit trickier” to add to the Young.AI arsenal, Galkin continues. The latter, referred to as the DeepMAge clock, will likely be the first of these to become available at Young.AI later this year, he adds.
Better Methylation Clock
Research on the epigenetics of aging and the novel DeepMAge clock, authored by scientists with Deep Longevity and Insilico Medicine, recently published in Aging and Disease (DOI: 10.14336/AD.2020.1202). The key capability of DeepMAge is to predict human age based on DNA methylation changes, making it a useful tool in aging research by letting scientists measure the effects of anti-aging interventions, says Galkin.
DeepMAge was trained to predict human age on more than 6,000 DNA methylation profiles. As described in a separate article in Ageing Research Reviews (DOI: 10.1016/j.arr.2020.101050), it can estimate human age within a 3-year error margin, which is more accurate than any other human aging clock, Galkin says. This includes the current industry standard—the 353 CpG clock, an age estimation method based on 353 epigenetic markers on DNA.
Unlike its predecessors, DeepMAge is a neural network and may additionally prove useful for the development of early diagnostic tools since it assigns a higher predicted age to people with various health-related conditions, says Galkin. For example, women with ovarian cancer and multiple sclerosis are on average predicted to be 1.7 years and 2.1 years older, respectively, than healthy women of the same chronological age. Similar results have been obtained for irritable bowel diseases, dementia, and obesity.
“We are currently looking into how DeepMAge works in tissue samples,” he says. “Surprisingly, although DeepMAge was trained only on blood samples, it can be repurposed for use in non-blood tissues.” The conundrum still being explored is why not all tissues work well with the aging clock.
“Right now, we are looking for genetic pathways that affect age prediction and can be affected by dietary supplements,” says Galkin. The two issues that need to be worked out are if tissue epigenetic landscapes can be approximated from blood epigenetics, and the connectedness of epigenetics and gene expression.
The second issue seems easy to resolve, he adds, but “keep in mind that we are looking for specific pathways and thus we need to select the genes we want to inspect carefully.”
In a follow-up study, DeepMAge could theoretically be tested with cells that have been epigenetically reprogrammed to a younger age—a feat that has recently been accomplished by researchers at Harvard Medical School.
The neural network architecture that DeepMAge is based on can be modified to digitally emulate the effects of lifestyle changes (e.g., fasting, taking longevity supplements, physical training), which is what makes this aging clock a likely next addition to Young.AI, says Galkin. Consumers and clinics are both targeted customers for the web platform.
At some point, the model will be released to the academic research community, either as a Young.AI app or as a separate web platform, Galkin adds.
Aging clocks have proven to be indispensable tools in aging research since 2013 after the first DNA methylation clocks were published in Molecular Cell (DOI: 10.1016/j.molcel.2012.10.016) and Genome Biology (DOI: 10.1186/gb-2013-14-10-r115).
DeepMAge is a “conventional multilayer perceptron,” explains Galkin, which is a feedforward neural network with fully connected hidden layers. It accepts a 1,000-long vector of methylation beta-values that get multiplied and summed in a convoluted way several times to produce a single number—the donor's biological age.
“We initially normalized all the data according to one specific protocol, but it turned out other normalization methods work just fine,” he says. “It makes using DeepMAge easier for scientists who wouldn't need to reproduce our data preprocessing protocol to use [the tool].” A patent application for DeepMAge has been submitted.
Before Young.AI users could utilize the DeepMAge or BloodAge clocks to assess their health or response to treatment, they would need to make an appointment with a physician to have their blood drawn and sent to a laboratory, Galkin says. Once Deep Longevity decides on a lab service provider for the methylation part, he adds, the user experience with DeepMAge will also be no different than for a routine blood test.
Eventually, Young.AI will have a $14 monthly fee for people who want to access extended functionality, says Galkin. “Some aging clocks will be available free of charge for all users, although the free versions may not provide such detailed reports as the paid versions.”
Physicians will gain access to medical-grade reports through planned partnerships with clinics, which may want to have the reports tailored to the needs of doctors and patients, he adds. Standard reports could also be available through paid API access.