Ab-Ag inverse folding methods benefit from pretraining

Summary

For inverse folding of antibody-antigen interfaces, pretraining on all monomeric protein structures prior to fine-tuning on antibody structures improves sequence recovery from 34.5% to 47.7% (1). General-protein pretraining particularly improved CDRH3 recovery, while antibody-specific fine-tuning gave the best overall performance. Pretraining on general protein-protein interaction structures before antibody-specific fine-tuning did not improve sequence recovery over antibody-only training.

See also

• Structure prediction and design tools trained on monomers generalize to oligomers

1.

Mahajan SP, Dávila-Hernández FA, Ruffolo JA, Gray JJ. How well do contextual protein encodings learn structure, function, and evolutionary context? Cell Systems. 2025;16(3):101201. Available from: https://doi.org/10.1016/j.cels.2025.101201