Polypharmacological Target-Aware Molecule Generation.

PolyTamGen, an autoregressive sequence-based framework, improves joint success rates in multi-target ligand generation for dual- and triple-target benchmarks, using a dataset of 1.6 million ligands.

• Why it matters: Designing compounds that modulate multiple proteins is essential for treating complex diseases like cancer and neurodegenerative disorders, but current AI methods are limited by scarce structural data and inefficient pipelines.
• What they did: Researchers curated a large-scale ligand-centric bioactivity dataset and developed PolyTamGen, which encodes target sets with ESM3, compresses features via an attention module, and generates multi-target ligands, further enhanced by Direct Preference Optimization.
• The result: PolyTamGen maintains high chemical validity and diversity, demonstrating improved joint success rates over baselines, thus enabling scalable multi-target drug design without relying on scarce structural priors.