Thermostability engineering of microbial transglutaminase using artificial intelligence and investigation of its underlying mechanisms

The application of artificial intelligence in enzyme molecular evolution has emerged as a research hotspot. However, applying machine learning to enzyme molecular modification still presents many challenges. In particular, accelerating the integration of machine learning and rational design is one of the important development trends in the field of protein engineering. In this study, we experimentally validated key amino acid mutations (E164L, E164P, S199E, and S199Q) predicted by a lab-developed sparse convolutional neural network to enhance the thermostability of microbial transglutaminase. We further investigated the molecular basis of this enhanced stability using molecular dynamics simulations. Compared with the wild type MTG, the thermal stability of the four mutants was significantly improved, and S199E showed the most remarkable improvement. At 60 °C and 50 °C, the half-lives of S199E were 2.3 times and 5.8 times those of the wild type, respectively, and the enzyme activity was increased by 1.4-fold. Molecular dynamics simulations showed that the binding free energy of S199E was - 28.68 kcal/mol, slightly lower than that of the wild type (- 27.96 kcal/mol). The root mean square deviation and root mean square fluctuation of the S199E mutant were 0.25 nm and 0.0566 nm, respectively, showing no significant changes compared with the wild type. LigPlot analysis indicated that E199 formed one hydrogen bonds with A309 and three salt bridges with H201, which might enhance local stability. These findings indicate that the improved thermal stability of the S199E mutant arises from enhanced local structural stability, not from major changes in overall protein structure, and accounts for its slightly lower binding free energy compared to the wild type.