Preference optimization of protein language models as a multi-objective binder design paradigm
This work addresses the challenge of multi-objective binder design for drug development, representing an incremental advance in applying language model alignment techniques to protein engineering.
The authors tackled the problem of designing protein binders with multiple objectives by fine-tuning a protein language model using direct preference optimization, resulting in generated binders showing median isoelectric point improvements of 17% to 60%.
We present a multi-objective binder design paradigm based on instruction fine-tuning and direct preference optimization (DPO) of autoregressive protein language models (pLMs). Multiple design objectives are encoded in the language model through direct optimization on expert curated preference sequence datasets comprising preferred and dispreferred distributions. We show the proposed alignment strategy enables ProtGPT2 to effectively design binders conditioned on specified receptors and a drug developability criterion. Generated binder samples demonstrate median isoelectric point (pI) improvements by $17\%-60\%$.