AI company Insilico selected drug candidates in an average of 13 months; animal toxicology studies take 9–12 months and human trials take years, says founder Alex Zhavoronkov

AI company Insilico selected drug candidates in an average of 13 months; animal toxicology studies take 9–12 months and human trials take years, says founder Alex Zhavoronkov

On November 3, Harvard geneticist David Sinclair published an interview with Zhavoronkov in Lifespan magazine. Insilico’s founder outlined the development timeline for rentosertib, an experimental drug for pulmonary fibrosis (scarring of the lungs): AI identified the target (the protein the drug acts on) and designed the molecule.

Insilico defines a drug candidate as a molecule that has undergone cell and animal testing to assess its absorption, activity against the disease, and toxicity in preliminary experiments. At that point, only the studies required by regulators remain before the first human dose. From 2021 to 2024, the company selected 22 such molecules, according to its own data, taking an average of about 13 months from project launch to candidate selection. Zhavoronkov says this process usually takes about 4,5 years.

Rentosertib progressed from target discovery to candidate selection in 18 months and entered its first human trial within 30 months; the computational work that produced the molecule took about three weeks. From there, Zhavoronkov says, biology set the pace: about five years elapsed between the project’s launch and positive phase IIa results from the first 12-week trial in 71 patients. In September 2026, the first patient received the drug in phase III, the final phase before approval. The trial in China plans to enroll 320 participants, with treatment lasting 52 weeks.

The trial’s primary outcome is the rate of decline over a year in the volume of air a person can forcibly exhale after a deep breath: pulmonary fibrosis causes irreversible deterioration in lung function. The trial leader estimates that, under favorable conditions, approval would come three to four years after the start of phase III. Insilico is pursuing drugs for aging through disease-specific development: it seeks approval to treat a disease while measuring aging alongside it. Zhavoronkov considers this the fastest realistic route.

Zhavoronkov’s main priority is to shorten the preclinical studies required by regulators and get trials started sooner. He estimates that a regulatory 28-day animal toxicology study takes 9–12 months once preparation, tissue analysis, and reporting are included, nearly as long as the entire candidate discovery process. If regulators more often accepted early trial designs informed by measurable biological indicators and AI, he says, those trials could proceed faster while maintaining quality.

He estimates that AI and automation could cut 2–3 years from a project involving a new target. He predicts that scientists in 2046 will direct laboratories staffed by AI agents and robots, choosing the research questions and taking responsibility for patient safety.

“Computation is now cheap. Demonstrating safety and efficacy in living systems still takes years.”

Zhavoronkov’s first concern is exaggerated promises in aging research. He recalls the words of Harvard Medical School dean George Daley at the opening of the ARDD conference on aging: “I worry that the optimistic claims made by parts of the aging research community far exceed what we can responsibly conclude from our current understanding of biology.” His second concern is a bubble in AI drug discovery: some private companies are valued at billions of dollars despite having no drugs in clinical trials, and if the bubble bursts, funding for rigorous work could dry up. His third concern is geopolitics: a major economic crisis or war could slow progress for an entire generation.