Inductive launched Beacon-2 , a system that predicts the efficacious human dose of a small molecule directly from its chemical structure.

Dose is the number that decides whether a compound can become a drug. It accounts for both how potent a molecule is and what the body does to it. A drug that works at a lower dose also carries less risk of off-target toxicity.

Beacon-2 predicts the properties that determine dose (absorption, distribution, metabolism, excretion, and toxicity) and estimates potency, then combines them through mechanistic models of pharmacokinetics. The ADMET models underneath it have won three consecutive OpenADMET blind challenges, beating more than 750 competitors from leading AI and pharma companies.

A technical post published today covers how Beacon-2 works and how its projections across 20 real-world drug programs and 325 publicly available compounds from the ExpansionRx OpenADMET competition.

To test how Beacon-2 impacts agentic design workflows, Inductive gave its medicinal chemistry agent Indy a recently disclosed SARS-CoV-2 compound for optimization. Over 5 autonomous design cycles, Indy improved predicted human dose by 17x, which can be the difference between a drug program that gets killed and one that makes it to clinical trials.

"Chemists have always had to balance a dozen separate properties against each other to decide whether a compound is worth making," said Josh Haimson, co-founder and CEO of Inductive. "Dose is what those properties all add up to. Computing it from structure changes which compounds a team decides to make and helps them identify the highest quality compounds faster."

Beacon-2 is available to partners through Compass, where it is already running on live drug discovery programs.