Fine-tuned antibody language model steers CDR mutations toward tighter SARS-CoV-2 spike binding
- Ab-Affinity is a protein language model fine-tuned to score antibody binding to a SARS-CoV-2 spike peptide, then used as the fitness function in a genetic algorithm with simulated annealing to mutate CDRs.
- Under cross-validation, held-out and independent tests it beat other language-model affinity predictors; its designs scored far higher than the parent antibody and than alternative mutation strategies.
- Support for top designs is mostly computational (docking, MD); one candidate showed apparently improved binding in vitro. Code and weights are released.
💡 Points Based on the abstract
One in vitro candidate cannot validate a design loop whose fitness function is the same model being reported as accurate. A GA with simulated annealing will exploit any blind spot in the scorer, so higher predicted affinity over the parent is expected by construction. Ask for SPR or BLI on a dozen designs, spanning high and mid scores, and a comparison against ESM-style likelihood and structure-based baselines.
📄 Paper · PLOS Computational Biology (PLOS) ↗ An antigen-specific protein language model guides CDR optimization for antibody design