Protein language models are better zero-shot predictors for ranking closely related sequences than distantly related sequences

Summary

Protein language models are better zero-shot fitness predictors for ranking closely related sequences than more distantly related sequences (1). This was shown by fine-tuning various sequence-only language models on mutations from mutants in one wildtype sequence, and inferring the effect of mutations from a different wildtype sequence.

Figures

Ref (1)

See also

• Antibody LMs outperform generic PLMs on intrafamily thermostability prediction

1.

Didi K, Alamdari S, Lu AX, Wittmann B, Johnston KE, Amini AP, et al. FLIP2: Expanding Protein Fitness Landscape Benchmarks for Real-World Machine Learning Applications. openRxiv; 2026. Available from: https://doi.org/10.64898/2026.02.23.707496