The abstract is read at the publisher; the summary is JClub's.
ProLigGPT, a GPT-2–based seq2seq language model trained on ~1.9 million protein–ligand pairs without 3D structures, achieves superior ligand-design performance across eight target proteins.
- Why it matters: Existing protein-targeted ligand models often rely on structural constraints, focus on single tasks, and struggle with cross-domain protein-to-ligand generation.
- What they did: ProLigGPT performs de novo ligand generation and scaffold-based design for target proteins, then is benchmarked against three structure-based models on eight representative targets.
- The result: It uses the full protein sequence to generate target-aware ligands, designs Pl3Kalpha inhibitors with novel and known scaffolds, and generalizes on Target T1 lacking crystal structure and known inhibitors.
The findingWhy it mattersWhat they didThe result
AI summary of the abstract · may contain errors · not medical advice · read the paper.