'Aging clocks' tell you how much 'older' you are than your chronological age. How do they work?

Think of a battery: New batteries start at 100% capacity, in terms of their ability to reliably hold a charge, but that capacity drops over time as the battery cycles on and off and powers devices. Biological age is a similar concept of capacity, and the tools researchers and clinicians use to measure your capacity are known as aging clocks, also known as "omic clocks" or "biological age tests."

Although such clocks are in development, the science of biological age is still in its infancy. The first descriptions of aging clocks appeared in journals in 2013. Since then, researchers have developed dozens of aging clocks that measure biological age via different metrics, such as protein profiles, immune-system function and epigenetic modifications, meaning alterations to DNA that change how genes work without changing DNA's underlying code.

Aging clocks are typically built upon machine learning models — statistical models that recognize patterns in data and make predictions based on those patterns. These models are based on a mathematical technique called regression, which seeks to predict the likelihood of events based on many variables and their relative importance to the prediction, known as "weights."

In simple terms, the models multiply each variable by its weight and add up all the weighted variables to get your probability. For example, a regression model predicting a person's risk of lung cancer might weight a history of smoking closer to 1, because it's very strongly correlated with lung cancer but would weight radon exposure lower than smoking because it's not as predictive of lung cancer risk.

These machine learning models used in aging clocks are trained on thousands of "biomarker" data points. Biomarkers are measurements of certain compounds, often but not always from blood samples, that act as a proxy measure for a condition or biological process. For example, higher-than-normal levels of C-reactive protein and white blood cell count usually mean the immune system is responding to an infection. Blood is such a good source of biomarkers because it circulates through the whole body and inevitably picks up signs of disease, Sun explained.

Clocks are also trained on the chronological ages and health statuses of the people providing samples to the dataset.

The algorithm analyzes these data and looks for patterns — machine learning's main strength — before coming up with a set of rules with which to interpret new data points that weren't included in the original dataset. In that way, it can make predictions about a given person's health, even if it doesn't "know" their age or health status. It can just go off of biomarkers and patterns pulled from the original data.

The first aging clocks, as well as many of their successors, are based on epigenetics — specifically, DNA methylation data. Methyl groups are molecules that latch onto certain sites within DNA, influencing whether the gene they're attached to is active.

What's key is that these sites can gain or lose methyl groups over time. Methylation patterns vary across the body, and research suggests they change in predictable ways with age. By analyzing these typical patterns, an epigenetic clock can estimate an individual's biological age. The difference between their actual age and the predicted age — called the age gap, or the "delta" — determines whether they're aging faster or slower than the healthy norm.

A 2024 study in the journal Epigenomics details four generations of epigenetic clocks:

First generation: Trained on only methylation data and measured only the delta, or the difference between chronological age and computed biological age. They can tell how much "older" or "younger" you look compared to a norm.

Second generation: Added data sets around mortality and health conditions to predict people's risk of early mortality or age-related conditions. An example of a second-generation clock is PhenoAge, which incorporates datasets with biomarkers measuring liver, kidney, metabolic and immune function. By adding these other data, PhenoAge can predict the risk of all-cause mortality, heart disease, cancer, Alzheimer's disease and more.

Third generation: Estimate both the age gap and how quickly or slowly someone is aging in terms of a rate. Whereas first-generation clocks are more of an odometer, tracking how far you've gone, these third-generation clocks are more like a speedometer, telling you how fast you'll get to where you're going. Examples include DunedinPACE and DunedinPACNI.

Fourth generation: Analyze specific methylation sites that are believed to cause some of the physiological breakdown we call aging. They incorporate an epigenetic analysis technique called Mendelian randomization, which tries to tease out cause and effect and determine whether methylation or de-methylation at certain sites are a cause or an outcome of age-related breakdown. This analysis enables these clocks to move beyond prediction and start determining the root causes of aging, their developers say.

Changes in DNA methylation and other epigenetic markers are hallmarks of aging, but there are many others. Thus, other types of aging clocks measure biomarkers of those hallmarks.

Proteomic clocks, for example, look for patterns in an individual's protein profile, usually based on blood samples. Because proteins are involved in nearly all disease processes and proteins are the target of nearly every pharmaceutical in existence, researchers think proteomic clocks could zero in on the actual drivers of aging, potentially uncovering new targets for intervention.

Metabolomic clocks measure and make predictions based on your profile of metabolites, which are byproducts of metabolism, the body's process of converting nutrients into energy. Collection techniques for metabolomic data are inexpensive and widely available, making these clocks useful for large-scale population studies.

Other clocks are based on transcriptomics, meaning they look at patterns of gene activation based on circulating RNA in the body. As a graduate student at Stanford University, Sun co-authored a 2024 study in the journal Nature about an algorithm that finds transcriptomic patterns related to age in brain cells.

Meanwhile, the DunedinPACNI clock is based on brain structure data gathered from MRIs. Some clocks are organ-specific, some are cell-specific, and some combine other clocks to create "multiomic" aging clocks.

For aging clocks to be useful, "they would need to be both prognostic — able to tell the future — and they would need to respond to interventions," said Dr. Dan Henderson, a primary care physician at Brigham and Women's Hospital and an instructor of medicine at Harvard University Medical School. In other words, clocks would need to accurately predict patients' risk of disease and shift in response to a person receiving effective treatment; if the treatment is working, one's "age" should go down.

For now, Sun thinks the most useful applications of aging clocks remain in the lab. He said that these tools could feasibly help determine if a treatment is actually affecting the aging process. Instead of following study subjects for years to see how a treatment affects their health outcomes, scientists can make reliable predictions based on samples taken before and shortly after treatment.