Rentosertib treats idiopathic pulmonary fibrosis, a disease where lung tissue scars and stiffens until breathing fails, usually within three to five years of diagnosis. Insilico's September 7, 2026 paper in Nature Biotechnology describes how the company found it: a generative model trained on protein structure data proposed a novel target protein linked to the scarring process, and a second model designed a small molecule to bind it, compressing a search that normally takes medicinal chemists years of trial-and-error synthesis into a few months of computational screening before anything touched a lab bench. The company has now published the pipeline itself, not just the drug, as what it calls a toolkit for other researchers to run the same two-model process on different diseases.

The distinction that matters here is between finding a molecule and finding a target worth hitting. Most AI-in-drug-discovery claims are about the first part, generating candidate compounds against an already-known protein. Insilico's system did the harder step too: it proposed the target itself, based on patterns in fibrosis-related protein data, which is closer to the actual bottleneck in drug development, since picking the wrong target is the reason most clinical trials fail regardless of how good the resulting molecule is. Rentosertib is still a single Phase 2a asset with results published, not an approved drug; the gate that decides whether the toolkit is more than a good paper is whether an outside lab, using Insilico's published method on a disease Insilico didn't choose, produces a second clinical candidate.