General-purpose language models underperform specialist models on tested epitope identification tasks

Summary

General-purpose text language models underperform matched specialist models on the tested epitope-identification tasks (1). This was carried out on known structures rather than prospectively using antibody design.

See also

• Antibody structure prediction
• Correct antibody-antigen prediction in AF3 and related models is partially determined by training set similarity

1.

Wang Z, Wang J, Wang Q, Xu Y, Du G, Hou T, et al. EpiBench: Can LLMs Understand Epitopes for Antibody Drug Discovery?. 2026. Available from: https://arxiv.org/abs/2608.06022