#31 · Structural bio & protein design

Generative and structure-based models guiding antibody and small-molecule binder design

The shared thread is coupling generative or predictive models to explicit structural, chemical, or assay context rather than using them as one-shot predictors. OFAntibody performs all-atom de novo design of monoclonal and bispecific antibodies with epitope conditioning. Hyper Screening X pairs a generative flow network encoding reaction logic with structure-based design to search an 11-trillion-compound make-on-demand space. The LasB study integrates fragment-based generative design with alchemical absolute binding free-energy ranking and experimental validation. kitAb applies AutoML over 59 interpretable Fv descriptors for assay-specific developability modeling. The Bits to Binders competition evaluated 12,000 AI-designed CD20-targeting minibinders in CAR-T cells, with 707 designs showing enrichment. What is new is the extension of context-aware generation to antibodies, minibinders, and small molecules, with prospective experimental checks. Evidence remains preliminary and largely preprint-based.