Deep learning-guided engineering of pectinase for enhanced catalytic performance in tobacco processing

Deep learning is revolutionizing enzyme engineering through efficient residue redesign. Leveraging deep learning for enzyme engineering, we redesigned a pectinase using ProteinMPNN guided by multiple sequence alignment. Our top-performing variant, DS-5, incorporated 72 mutations and achieved an 8.9-fold increase in catalytic activity compared to the wild-type. DS-5 also displayed significantly improved thermostability, with an optimal temperature increasing by 10°C, and robust performance over a wide pH range (7.0-11.0). Structural and molecular dynamics analyses revealed the source of this enhancement: a remodeled surface electrostatic potential due to the increase of five positively charged residues, forming an extended positive groove that potentially improves substrate binding affinity. This rationally designed enzyme demonstrated superior performance in applied settings, including apple juice clarification and tobacco degradation. Furthermore, treating tobacco leaves with DS-5 substantially improved their sensory profile by elevating the concentration of desirable flavor compounds like sucrose and lactones. Our study provides a framework for deep learning-guided engineering of highly efficient enzymes, directly linking catalytic improvements to enhanced end-product quality for industrial applications.