Insilico CEO Alex Zhavoronkov calls doubling human lifespan with AI an ambition and separates biological age measured from blood proteins from a longer life

Insilico CEO Alex Zhavoronkov calls doubling human lifespan with AI an ambition and separates biological age measured from blood proteins from a longer life

On October 14, Zhavoronkov, who heads the biotech company Insilico, published a column in Fast Company.

"I believe AI may eventually help double the human lifespan. But that is an ambition, not a forecast we can test today."

He writes that drug development moves at two speeds. The computational part is fast: AI suggests drug targets and designs molecules. The biological part works differently. A molecule has to be synthesized and tested, and in human trials patients have to be enrolled and the results awaited. His example is rentosertib, a drug for pulmonary fibrosis (scarring of the lungs) that Insilico designed with AI. Going from the target to a candidate molecule ready for laboratory testing took about 18 months. After that, a phase IIa trial in 71 patients gave what Zhavoronkov describes as an encouraging signal on lung function, and phase III is now underway. AI can also speed up the trials themselves, but one question is especially hard to answer: whether a preventive treatment will help a healthy person decades later.

Protein clocks estimate biological age from proteins in the blood. In a paper by Zhavoronkov's group in Nature Biotechnology, published September 7, six such models were applied to blood from 42 of the 71 phase IIa participants. In patients who received the drug, all six models showed an age lower than at baseline. At 60 mg once daily, four of the models predicted a larger age reduction than with placebo by week four, by 2.7–3.5 years. Zhavoronkov writes: "This does not mean the patients literally became years younger." The analysis is preliminary, the follow-up was short, and the change in proteins may reflect several processes, including ones related to the disease.

One of the paper's reviewers described this problem. The clocks are built on proteins whose levels distinguish young people from old ones, yet many of these same proteins also distinguish sick people from healthy people of the same age. The reviewer therefore expected that successfully treating a disease would shift these protein levels in a way that makes the clocks show "rejuvenation", even if the aging processes themselves are untouched. The authors point to an indirect argument against the idea that the clocks simply track the easing of the disease: lung function improved most at 60 mg once daily, while the clock shift agreed most consistently across the models under a different regimen, 30 mg twice daily.

In their response to the reviewers, signed by Zhavoronkov, the authors reframed the work "from a report of clinical age reversal to the first implementation of a clinical trial design template", meaning a model for building protein clocks into trials of drugs for specific diseases. In an Insilico press release dated September 10, rentosertib "demonstrated the potential to reverse biological age", while the column describes the same result as "treatment-associated shifts toward younger predicted profiles".

Zhavoronkov advises people buying biological age tests, and investors, to find out what a change in such an age actually predicts (better physical condition, fewer diseases, or a longer life) and whether that link has been demonstrated for the specific intervention being tested. Companies, he writes, should decide in advance which people to test on, what counts as success and what evidence is needed. Investors should look at whether a company can repeatedly turn predictions into reliable experiments and clinical data: "The number of molecules a model generates matters less than the quality of the decisions that follow."