Closely related protein language models have diverging performance on specific domains

Summary

Closely related protein language models can have diverging performance on specific domains, even when the architecture and training datasets are the same (1).

Figures

Ref (1)

1.

Dinh T, Jang S-K, Zaitlen N, Ntranos V. Compressing the collective knowledge of ESM into a single protein language model. Nature Methods. 2026; Available from: https://doi.org/10.1038/s41592-026-03050-9