Reproduce “Emergent Biological Capabilities in a Foundation Model for Molecular Interactions” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Adi Mashiach
Bonbon uses paired protein and ligand sequences in a transformer to predict interaction properties from a shared frozen representation, without structural coordinates or task-specific labels in pretraining. Its predictions cover mechanism of action, binding-site location, and covalent engagement. Six-class mechanism-of-action macro precision rose from 0.608 at 1.3B parameters to 0.811 at 1.6B; one run per scale leaves the cause unknown.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.