Prediction of Enzyme function using interpretable optimized Ensemble learning framework

Abstract

Accurate prediction of enzyme function, particularly for newly discovered uncharacterized sequences, is immensely important for modern biology research. Recently machine learning (ML) based methods have shown promises. However, such tools often suffer from complexity in feature extraction, interpretability, and generalization ability. Here we present an interpretable ML method, SOLVE that addresses these issues by using only combination of tokenized subsequences from the proteins primary sequence for classification. Its optimized ensemble learning framework improves prediction accuracy, distinguishes enzymes from non-enzymes, and predicts enzyme commission (EC) numbers for mono- and multi-functional enzymes. Additionally, SOLVE provides interpretability through Shapley analyses, identifying functional motifs at catalytic and allosteric sites of enzymes. By focusing only on primary sequence data, SOLVE simplifies high-throughput enzyme function prediction for functionally uncharacterized sequences and outperforms existing tools. With high prediction accuracy and its identification ability of functional regions, SOLVE can become a promising tool in different fields of biology and therapeutic drug design.