SeqSol-FECS: An Integrated Deep Learning and Structural Modeling Framework for Protein Solubility Engineering

Abstract Escherichia coli is widely used as a workhorse for heterologous protein expression. However, insoluble protein expression is frequently encountered. The mechanisms underlying protein solubility remain incompletely understood, limiting the effectiveness of current protein solubility engineering strategies. Here, SeqSol, a sequence-level solubility prediction model capable of prioritizing single-amino-acid substitutions through WT−mutant rescoring, was developed using 71,425 proteins. SeqSol achieved an accuracy of 75.8% and an AUC of 0.840 on an independent test set and correctly identified experimentally validated solubility-enhancing mutations from diverse proteins. To minimize the risk of activity loss, candidate mutation sites were prioritized according to conformational flexibility, relative solvent accessibility, and evolutionary conservation before solubility engineering. Using only 30 experimentally tested mutations, the resulting SeqSol-FECS framework increased the solubility of two industrial enzymes from 56.18% and 10.53% to over 95%, respectively, while maintaining their specific activities. SeqSol-FECS provides an efficient strategy for enzyme solubility engineering and biomanufacturing applications.