Enhancing prediction accuracy for enzyme activity engineering through sequence co-evolution and epistatic relationship modeling

Abstract

Abstract Conventional enzyme engineering strategies, such as directed evolution with structure-based analysis, are limited by laborious workflows and by the need to screen extensive enzyme libraries. Computational models based on multiple sequence alignment, such as SCANEER, have streamlined these engineering strategies, enabling the successful prediction of single-site mutations that enhance enzyme activity. However, when combining predicted mutations to improve enzyme performance or expand mutational space, such approaches often fail due to epistatic interactions among amino acid residues, which can lead to complex, non-additive effects. Here, we present epiSCANEER, an extended computational framework that incorporates co-evolutionary dependencies reflecting residue-level epistatic effects to identify mutation combinations with a higher likelihood of enhancing enzyme activity. By evaluating amino acid pairs at co-evolved positions across homologous sequences, epiSCANEER prioritizes mutation combinations with high evolutionary compatibility, significantly narrowing the combinatorial search space to variants more likely to exhibit activity-enhancing effects. Experimental validation demonstrated a 76% success rate for epiSCANEER prediction, compared to 30% and 38% for single-site predictions and combinatorial approaches of single-site mutants, respectively. This novel method obviates the need to construct single-mutation libraries, significantly reducing labor and costs while improving success rates. epiSCANEER has been developed as a web-server that enables researchers to access tools for rational enzyme optimization.