Abstract
Thermostability is a critical goal in protein engineering for applications of biocatalysts and biomedicines. Despite striking advances in biomolecular predictive modeling, reliably identifying stabilizing mutations remains challenging. Previously, molecular dynamics (MD) simulations and visual inspection have been used as a secondary filter to improve the success rate of mutations pre-selected by thermostability algorithms. However, this approach suffers from low throughput and subjectivity. Here, we introduce Biophysical Overview of Optimal Stabilizing Mutations (BoostMut), a computational tool that standardizes and automates mutation filtering by analyzing dynamic structural features from MD. BoostMut formalizes the principles guiding manual verification, providing a consistent and reproducible stability assessment. Rigorous benchmarking across multiple datasets showed that integrating BoostMut's biophysical analysis improves the prediction rate regardless of the initial thermostability predictor. Given a modest amount of existing mutant stability data, BoostMut's performance can be further enhanced with a lightweight machine learning model. Upon experimentally validating BoostMut predictions on the enzyme limonene epoxide hydrolase, we identified stabilizing mutations previously overlooked by visual inspection and achieved a higher overall success rate. We foresee BoostMut being used for mutation filtering, as an integrated step in thermostability prediction workflows, and for labeling data to train future predictors.