Evaluating Machine Learning Models on Predicting Change in Enzyme Thermostability

Abstract

Enzymes are efficient catalysts for biological reactions and can potentially be designed to speed up non-biological reactions, such as reactions in industrial processes. However, physically experimenting with new protein designs is time consuming, and an efficient method to predict protein stability is needed. Our research problem is finding the best machine learning model to predict the change in enzyme thermostability after a single point mutation in the amino acid sequence. We trained several machine learning models and found that the XGBoost model had the best performance with an R2 score of 0.593 (R2 score is a metric where higher is better and a perfect model would have a score of 1).