Reproduce “iGC: Antibody Optimization via Reinforcement Learning over a Learned Somatic Hypermutation Prior” 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
Thao Nguyen, Robert G. Alberstein, Ewa Nowara, Homa MohammadiPeyhani, Holly R. Steach, Sarah A. Robinson, Yuval Shechter, Donald W. Lee and 5 more
iGC learns mutation proposals from lineages, then optimizes them toward target binding while regularizing the policy toward its learned mutation pattern. On a held-out test, it predicts mutation sites at 11.929%, versus 5.307% for the baseline; in GCReplay, its Top 100 reward–diversity product is 150.9 versus 32.1 for RewardNet alone. Rewards use single-substitution measurements, not testing of generated candidates.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.