Direct discovery · Chemistry · Medicine

Chemical language model yields nanomolar A2A receptor ligands

Experimentally confirmed de novo adenosine A2A receptor ligands, including three nanomolar ligands and two novel chemotypes characterized by co-crystal structures.

Summary

A peer-reviewed study combined a chemical language model with reinforcement learning and structure-based scoring to generate previously non-commercial A2A receptor ligands. Experimental testing reported an 88% binding hit rate and 50% functional activity, with co-crystal structures for the strongest binders.

AI role

A generative chemical language model and reinforcement-learning optimization generated and prioritized molecules from receptor structure without prior ligand chemistry.

Narrative role

This adds experimentally grounded evidence that generative molecular models can enter receptor-structure-guided chemical space and produce validated drug-like starting points, distinct from the tracker’s multi-target compound example.

Caveat

The result is a target-specific early discovery demonstration; receptor binding and in-vitro activity do not establish efficacy, safety, or clinical utility.