Longevity Briefs: Using AI To Find Untapped Anti-Ageing Drugs

Longevity briefs provides a short summary of novel research in biology, medicine, or biotechnology that caught the attention of our researchers in Oxford, due to its potential to improve our health, wellbeing, and longevity.

The problem:

Biological age is a measure of how “old” your body’s systems are compared to your chronological age (the number of years you have been alive). Though imperfect, scientists can estimate biological age using “ageing clocks” that read certain molecular patterns (most commonly epigenetic modifications that alter how the DNA is read). One of the main uses of such clocks is to provide an idea of whether a given intervention slows ageing over a short period of time. If someone takes drug x for a year and their biological age only advances by a few months, this could indicate that they are ageing at a slower pace.

A lot of research unrelated to the study of ageing still collects data that could be used to estimate biological age, but isn’t, because it’s not what the researchers were trying to investigate. This means that there’s a large amount of data just waiting to be analysed, with molecules able to affect the biology of ageing potentially hidden in plain sight. Sifting through such vast datasets would be an impossible task for humans, but not for machine learning algorithms. In this study, which is still in preprint and not yet peer reviewed, researchers develop such an algorithm in order to mine public datasets and discover potential drugs of interest.

The discovery:

The researchers built ClockBase Agent, a three-part system designed to process over 2 million public human and mouse molecular samples, including epigenetic and RNA sequencing data. Rather than simply searching for compounds with an effect on biological age, ClockBase was designed to act somewhat like a real scientist, formulating hypotheses and evaluating them based on the available data and literature using appropriate statistical techniques. Researchers used ClockBase to apply over 40 ageing clock models to the data, flagging thousands of age-modifying effects missed by original investigators. These included over 500 interventions that were associated with significantly reduced biological age, as well as many interventions that accelerated ageing.

Encouragingly, ClockBase’s findings agreed with existing evidence. For example, ClockBase pointed to the immune-modulating drug rapamycin and the diabetes drug metformin, both of which have already been shown to lower biological age measurements. However, ClockBase also offered some new leads such as ouabain (a treatment for low blood pressure at low doses) and fenofibrate (sometimes used to treat abnormal blood lipids). The researchers were also able to independently verify the effects of ouabain in their own mouse study, reporting that it was able to reduce frailty, improve cardiac function and reduce inflammation in the brains of aged mice.

In addition to these drugs, ClockBase also identified some environmental factors associated with biological age measurements. These included ‘mechanical overload’ (as a result of exercise), which was associated with reduced biological age, as well as exposure to viruses, which had the opposite relationship with biological age.

The implications:

ClockBase is an example of how AI tools can be put to use to identify new avenues of research at little cost. Vast amounts of untapped scientific data exist for all fields of science, so why not put it to good use? The researchers found that ClockBase did make some reasoning errors when it came to interpreting some of the more complex experiments, which is why it is necessary to validate its findings. The fact that the researchers’ ouabain study appeared to agree with the AI suggests that tools like ClockBase are at least capable of providing promising leads.