Discrimination of acidic and alkaline alpha/beta hydrolase enzymes through data-driven machine learning approaches

Alpha/beta-hydrolases form a broad and versatile enzyme superfamily that includes lipases, esterases, carboxylesterases, and amidases key catalysts in both cellular metabolism and industrial biotechnology. Although they share a similar α/β fold and catalytic triad, members of this family often exhibit very low sequence identity, which makes it challenging to classify them solely based on sequence similarity. This challenge becomes even greater in extremophilic hydrolases that have adapted to survive and function in very acidic or alkaline environments. In this study, we aimed to classify α/β-hydrolase enzymes originating from acidophilic and alkaliphilic microorganisms by using a machine learning approach. A total of 403 bacterial enzymes were analysed based on their amino acid composition and physicochemical characteristics, without relying on secondary structure information. Several supervised learning models were tested, including Decision Tree, Random Forest, eXtreme Gradient Boosting, and Support Vector Machine. Among them, the Random Forest model showed the highest performance, reaching 76% accuracy with an AUC value of 0.90 ± 0.03. Alanine, lysine, and the grand average of hydropathicity (GRAVY) emerged as the most influential features in distinguishing acidic and alkaline enzymes. Overall, the results suggest that integrating sequence-derived and chemical descriptors through machine learning can effectively separate acidic from alkaline α/β-hydrolases. The findings not only shed light on molecular adaptation mechanisms to extreme pH but also provide a computational basis for guiding the discovery and engineering of new extremozymes for industrial applications such as bioremediation, detergent design, and biomining.