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 zero-shot capabilities in predicting molecular function, identifying binding sites, and distinguishing between covalent and non-covalent interactions, even for out-of-distribution complexes. Controlled scaling experiments revealed a sharp jump in these capabilities at the largest model scale, suggesting that traditional loss-based scaling laws are insufficient to predict emergent abilities. Prospective wet-lab validation demonstrated a 61.5% dose-response hit rate on de novo compounds for nine disease targets, yielding novel molecules. AI
IMPACT This model demonstrates emergent capabilities in molecular biology, potentially accelerating drug discovery and the development of novel therapeutics.
RANK_REASON The cluster contains a research paper detailing a new foundation model for molecular interactions. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
