First Human Test of AI-Created Medicine Helps People Reverse Aging

Generative AI was applied by Insilico Medicine to discover a new drug rentosertib for treating idiopathic pulmonary fibrosis (IPF), a progressive lung disease characterized by scarring of the lung tissue. During the clinical trial involving 42 patients, blood protein profiles were measured to track changes in their biological age at the same time.

1Six Algorithms, One Result

Biological clocks are machine learning algorithms that estimate a person’s biological age based on a panel of proteins found in their blood. Here, we employed 6 independent biological clocks developed by various groups including Harvard, Oxford, Beijing, and Insilico. These biological clocks were neither trained on the same dataset nor the same set of biomarkers. They all indicate that blood of rentosertib-treated patients was biologically younger than that of the placebo group by 3 to 6 years. One of the biological clocks even suggests that rentosertib could slow down aging by up to 6 years.

“I am impressed not so much by the size of the effect, but by the consistency among the clocks, since they were not trained on the same features and data”, says Nobel prize winner Michael Levitt. The study has been published in Nature Biotechnology.

2What This Means — and What We Don't Know Yet

Authors are careful with their conclusions. Decreasing “biological age” in blood does not necessarily mean that the patients became physically younger. The observed protein signature could resemble that of younger individuals because their lungs work better and the rest of the body is less stressed. However, the optimal dose for IPF treatment (60 mg/day) was different from the optimal dose for decreasing biological age (30 mg bid), hinting at some mechanism of action independent of lung health.

“This is promising, but we still need more data to make a final judgement. The main caveat is that the clocks were trained on diseased patients”, comments cardiologist Eric Topol. According to Harvard’s Vadim Gladyshev, “This study is the first to demonstrate clearly that predicted biological age can be lowered”.

3How AI Developed the Drug

Insilico had 2 AI tools. First, it searched through medical databases and scientific publications to find relevant proteins linked to the disease. Second, it examined the structure of these proteins and came up with potential candidates. The whole process from the target protein TNIK to a viable candidate molecule took approximately 18 months. Currently, rentosertib is in the last stage of clinical trials (Phase 3). Recently, Eli Lilly agreed to invest in Insilico, hoping to commercialize AI-generated drugs.