Overcoming Antagonistic epistasis in DAAO Engineering via Mechanism-Guided Multidimensional Feature Analysis

Abstract

Antagonistic epistasis often limits enzyme engineering by causing activity loss when beneficial mutations are combined. Here, we present a mechanism-guided multidimensional feature analysis (MDFA) strategy integrating rational design, structural partitioning, machine learning, and computational simulations to optimize D-amino acid oxidase (DAAO) for D-phosphinothricin (D-PPT). Channel geometry and electrostatics defined preferred screening ranges, while a CNN ensemble with a Random Forest surrogate enabled multisite classification. Computational and experimental analyses suggested that excessive local positive charge and imbalanced flexibility contribute to A. epistasis, whereas spatial partitioning reduces local conflict. OAEMT (N53R-V57R-S233 K-Q339R) increased catalytic efficiency 80.94-fold and converted 85.5% of D-PPT within 5 h in a 2 L reactor. MDFA thus improves combinatorial design within a defined mutational space.