Researchers have developed Bonbon, a foundation model for molecular interactions trained on approximately 3.2 trillion tokens of paired protein-ligand sequences. This model exhibits emergent capabilities in predicting molecular mechanisms of action, identifying binding sites, and distinguishing bond types, all without task-specific supervision or structural coordinates. Scaling the model from 42 million to 1.6 billion parameters revealed sharp increases in these capabilities at the largest scale, demonstrating that loss-based scaling laws are insufficient predictors. Bonbon achieved a 61.5% dose-response hit rate in prospective wet-lab validation for de novo compounds across nine disease targets, yielding structurally novel molecules. AI
IMPACT Demonstrates emergent capabilities in biological modeling, potentially accelerating drug discovery and molecular interaction research.
RANK_REASON Research paper detailing a new foundation model for molecular interactions. [lever_c_demoted from research: ic=1 ai=1.0]
Read on Hugging Face Daily Papers →
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →