Summary
Designs of the CDRH3 loops of Trastuzumab are more natural when designed by models fine-tuned from a general inverse folding model on structures of antibodies versus no fine tuning (1). Naturalness was judged using a CNN trained by (2). Similar performance can be achieved by ensembling logits from an antibody-specific protein language model with those of the inverse folding model without fine-tuning.
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
2.
Mason DM, Friedensohn S, Weber CR, Jordi C, Wagner B, Meng SM, et al. Optimization of therapeutic antibodies by predicting antigen specificity from antibody sequence via deep learning. Nature Biomedical Engineering. 2021;5(6):600–12. Available from: https://doi.org/10.1038/s41551-021-00699-9