Identification of nanomolar adenosine A2A receptor ligands using reinforcement learning and structure-based drug design
Problem
Finding novel small-molecule ligands for the adenosine A2A receptor from protein structure alone, without prior knowledge of ligand chemistry.
Why it matters
The A2A receptor is an important drug target, and new ligands or chemotypes could seed future medicines.
What the AI did
Generative chemical language models combined with structure-based drug design and Augmented Hill-Climb reinforcement learning generated and multi-objectively optimised novel molecules for protein-ligand complementarity.
Results
Experimental binding hit rate of 88%; 50% with confirmed functional activity; three nanomolar ligands; two novel chemotypes; two ligands co-crystallised with the receptor
Impact
Peer-reviewed work with experimental validation, including crystal structures confirming binding mechanisms of the generated ligands.
Record
| Venue | Nature Communications |
|---|---|
| Published | 2025 |
| Work type | article |
| Authors | Morgan C. Thomas, Pierre G. Matricon, Robert F. Gillespie, Maja Napiórkowska, Hannah Neale, Jonathan S. Mason, Jason Brown, Kaan Harwood et al. |
| Citations | 24 (OpenAlex, at import time) |
| References | 86 |
| Model or method | Generative chemical language models with Augmented Hill-Climb |
| Organization | not stated in source |
| Primary source | https://doi.org/10.1038/s41467-025-60629-0 |
| Open access | gold |
| Keywords | drug design, reinforcement learning, A2A receptor, generative models, nanomolar ligands |
Importance rating
significant — thousands to millions affected
Going from protein structure to novel ligands without prior chemistry is a genuinely new capability for drug design; abstract cites 88% hit rate and novel chemotypes for one GPCR target, no population scale.