Summary
PLMs make equally effective predictions when trained on individual proteins or protein families (1). Below, METL-L is trained on individual proteins and METL-G is trained on global sequence data.
Figures
Ref (1)
See also
- MD potentials from ML are more effective when protein-specific
- Antibody LMs outperform generic PLMs on intrafamily thermostability prediction
- Base PLMs must usually be fine-tuned to generate functionally active sequences
1.
Gelman S, Johnson B, Freschlin CR, Sharma A, D’Costa S, Peters J, et al. Biophysics-based protein language models for protein engineering. Nature Methods. 2025;22(9):1868–79. Available from: https://doi.org/10.1038/s41592-025-02776-2