Inductive's AI models place first among 270-plus entrants in OpenADMET blind test of drug-interaction risk
Its Beacon models predicted CYP inhibition potency for 750 withheld compounds across four enzymes, ahead of NVIDIA (third) and Recursion (fourth); it is the company's fourth straight OpenADMET win.
Why it matters
Cytochrome P450 enzymes metabolise most small-molecule drugs, and compounds that inhibit them can cause dangerous drug-drug interactions, so flagging that risk early lets chemists steer away from it. Blind challenges such as OpenADMET score models on compounds whose measurements stay hidden until the contest closes, which makes results harder to game than published benchmarks. Inductive credits its pooled ADMET data consortium for the win. The open question is whether benchmark leads like this translate into fewer failures in real drug programmes.
Key facts
- First of more than 270 eligible participants
- 750 compounds; CYP3A4, CYP2D6, CYP2C9 and CYP1A2
- NVIDIA placed third and Recursion fourth
- Fourth consecutive OpenADMET first place
- Beacon models tested on 19 endpoints across four challenges
- Announced in New York on November 6, 2026