Machine Learning Clock Predicts Lifespan

Two new machine learning clocks can accurately predict age, lifespan, and the effect of lifespan-extending interventions in mice

Let’s talk about age

Aging, or as we’ll consider it here, the period after reaching sexual maturity, is marked by the ‘gradual deterioration of functional characteristics’.

(Growth is, of course, also a process related to the passing of time and age, but that process is not what we are interested in here.)

The key word above is deterioration. While functional characteristics can change according to different life stages, what sets aging apart is the decline of function — even for functions suited to the stage of life.

For most people, it stills feels unconventional to call aging a disease. The case, however, has been made a few times by now. Whether you agree or not, it suffices that we recognize that, in the vast majority of people, advanced age is a period marked by various health problems.

(If you follow me in calling age a disease, these various problems can actually be thought of as ‘symptoms’ of an underlying problem, i.e. age.)

That point is that aging is a systemic process that leaves no bodily function unaffected. It is, to use a few extra syllables, a multi-factorial problem. And, as we noticed in the discussion concerning AI for mental health, machine learning/AI systems are particularly suited to deal with such issues. In fact, in a previous post, we looked at implementing machine learning in aging research. In that post, one of the possibilities we discussed involved the identification of biomarkers, certain ‘signals’ that could tell us our chronological age, as well as our potential lifespan.

Tik tok, time’s up

Now, a new study makes me feel a little bit like an oracle.

After all, in the work, researchers present two ‘clocks’ based on machine learning. One is a strong predictor of chronological age, and the other one accurately predicts lifespan and the effect of potential lifespan-extending interventions.

To develop these clocks, the researchers followed 60 mice from age 21 months to their natural death. (Although a lab is not really ‘natural’, you get the idea).

They tracked the frailty index (FI) of the mice, which is a composite of various parameters, ranging from coat color, over reflexes, to decline in vision, used to assess overall health.

Then, machine learning was unleashed on all that data. This resulted in two ‘clocks’ based on specific FI parameters: FRIGHT (Frailty Inferred Geriatric Health Timeline) and AFRAID (Analysis of Frailty and Death).

FRIGHT was a predictor of chronological age, i.e. without knowing the history of an individual mouse, FRIGHT can tell you how old it is.

AFRAID predicted life expectancy and the effect of interventions aimed at extending it.

To test the clocks, the researchers first looked at the data from a previous study in which mice were exposed to enalapril. This compound reduced FI score (aka better health), but does not affect lifespan.

Result: FRIGHT measured the mice as ‘younger’, but AFRAID didn’t budge. Exactly what you would expect.

Next, the researchers put mice on a methionine-restricted diet, known to extend lifespan.

Result: FRIGHT assessed them as younger, and AFRAID predicted they would live longer, which they did. Again, what we would expect if the two clocks were accurate.

Or:

These analyses demonstrate that the FRIGHT age and AFRAID clock models are responsive to healthspan and lifespan-extending interventions.

The clocks are not perfect (median error for FRIGHT is 1.3 months, and for AFRAID it is 1.7 months, both of which are substantial compared to a mouse’s lifespan of two to three years). Also, the mice in the study were all male, and all were C57BL/6 mice (a specific strain of lab mice, pictured above).

The clocks, in other words, can miss a tik or a tok here and there. However, at least we have clocks now. And, since knowledge is power, these clocks — and their future versions — will be powerful tools to help us truly understand the intricacies of aging, and, who knows, maybe treat it.

Future studies could develop a models based on the frailty items assessed here but modeled to predict a composite outcome including physiological measures in addition to chronological age. Still, even after the development of such composite clocks, the metrics described here — FI, FRIGHT age, and the AFRAID clock — will serve as rapid, non-invasive means to assess biological age and life expectancy, accelerating and augmenting studies to identify interventions that improve healthspan and lifespan.

Stay healthy and live long, all.