Researchers introduce DrugGen-2, a generative model designed to create small molecules conditioned on both disease ontology and target protein sequences. This approach addresses the gap in current computational methods by integrating specific disease context into the molecular generation process.
- Developed by fine-tuning a pre-trained GPT-2 model using supervised fine-tuning followed by reinforcement learning via group relative policy optimization (GRPO).
- Optimized with reward functions for chemical validity, novelty, diversity, and high predicted binding affinity.
- Evaluated on five protein targets relevant to diabetic nephropathy, significantly outperforming baselines like DrugGPT and DrugGen.
- Demonstrated superior capacity for unique molecules, greater structural similarity to approved drugs, and improved predicted binding affinities.
- Molecular docking identified candidate ligands with predicted affinities (-9.917, -9.485, -9.367) exceeding reference drugs like enalapril (-8.283).
By accounting for the complex interplay between diseases and molecular targets, DrugGen-2 advances AI-assisted drug discovery as a tool for de novo design and drug repurposing.