EITLEM-Kinetics: A deep-learning framework for kinetic parameter prediction of mutant enzymes

Abstract

The core issue in implementing in silico enzyme screening lies in accurately evaluating the merits of mutants. The best solution to this problem would undoubtedly be the precise prediction of kinetic parameters for mutant enzymes to directly assess the catalytic efficiency and activity of enzymes. Previously developed models of this type are mostly limited to predictions for wild-type enzymes and tend to exhibit poorer generalization capabilities. Here, a novel deep-learning model framework and an ensemble iterative transfer learning strategy for enzyme mutant kinetics parameter (k_cat, K_m, and KK_m) prediction (EITLEM-Kinetics) were developed. This approach is designed to overcome the limitations imposed by sparse training samples on the model’s predictive performance and accurately predict the kinetic parameters of various mutants. This development is set to provide significant assistance in future endeavors to construct virtual screening methods aimed at enhancing enzyme activity and offer innovative solutions for researchers grappling with similar challenges.