Prediction of Distal Mutation Effects in Enzymes via Integration of Molecular Dynamics Descriptors and Zero-Shot Model.
TL;DR
This work presents an integrated framework that combines molecular dynamics-derived descriptors with the zero-shot prediction model GEMS to identify beneficial distal mutations, offering an efficient and generalizable strategy for enzyme engineering.
Abstract
Identifying beneficial distal mutations remains a key challenge in enzyme engineering, as such residues can regulate activity through long-range dynamical coupling and allosteric communication. Although protein language models effectively capture sequence and evolutionary information, they lack explicit representation of conformational dynamics in the presence of the substrate, limiting their ability to detect distal regulatory sites. Here, we present an integrated framework that combines molecular dynamics (MD)-derived descriptors with the zero-shot prediction model GEMS to identify beneficial distal mutations. Compared to the pure zero-shot model, MD-derived descriptors efficiently capture key distal residues involved in dynamical coupling and allosteric communication with the active site. Consequently, these constraints enable the zero-shot model to predict distal mutations more precisely. By integrating sequence, evolutionary, and dynamic information, our approach expands the diversity of candidate sites while maintaining a manageable screening scale, offering an efficient and generalizable strategy for enzyme engineering.