Summary

Protein language models can be steered to design proteins with specific properties with greater success than traditional fine-tuning (1). The procedure involves training linear binary classifier heads on top of the activations of each layer and selecting the head with the highest validation score. Larger language models are more steerable, which contrasts with evidence against the scaling hypothesis of sequence-only language-only models in other domains. This was done with the ESM2-650M and ESM2-3B models.

Figures

Base ModelMethodThermostability ↑Thermostability Diversity ↑Thermostability Novelty ↑Solubility ↑Solubility Diversity ↑Solubility Novelty ↑
ESM2-650MOriginal Model56.48 (12.04)0.954 (0.023)0.591 (0.110)0.289 (0.151)0.963 (0.019)0.596 (0.130)
Fine-tuning63.56 (14.87)0.953 (0.023)0.585 (0.099)0.356 (0.213)0.961 (0.020)0.594 (0.132)
Activation Steering82.20 (12.92)0.971 (0.023)0.739 (0.130)0.494 (0.241)0.998 (0.001)0.880 (0.087)
ESM2 3BOriginal Model56.05 (11.24)0.968 (0.020)0.632 (0.143)0.298 (0.174)0.971 (0.021)0.622 (0.162)
Fine-tuning64.22 (14.49)0.965 (0.022)0.629 (0.143)0.385 (0.236)0.966 (0.022)0.610 (0.165)
Activation Steering83.33 (9.47)0.990 (0.006)0.915 (0.105)0.631 (0.228)0.996 (0.003)0.951 (0.077)

Ref (1)

1.
Huang L-K, Zhu R, He B, Yao J. Steering Protein Language Models. In: International Conference on Machine Learning. PMLR; 2025. p. 26247–60. Available from: https://proceedings.mlr.press/v267/huang25ba.html