Direct discovery · Chemistry · Medicine · Biology
POLYGON generates and tests dual-target drug-like compounds
De novo compounds designed for dual MEK1 and mTOR activity and tested in biochemical and cell assays.
Summary
The POLYGON study used generative reinforcement learning to design polypharmacology compounds for protein-target pairs, then synthesized 32 MEK1/mTOR candidates. The reported experiments found that most reduced both protein activity and cell viability at micromolar doses.
AI role
A generative reinforcement-learning model sampled molecular structures rewarded for predicted dual-target inhibition, drug-likeness, and synthesizability.
Narrative role
This backfills a non-duplicate 2024 small-molecule case where generative chemistry produced synthesized and experimentally tested compounds, strengthening direct-discovery coverage beyond proteins and materials.
Caveat
The evidence is early-stage biochemical and cell-assay activity, not animal efficacy, clinical validation, or proof that the compounds are optimized drug leads.