Physical Chemistry Seminar: Accelerating enzymatic reactions via sequence-based modeling: application to haloalkane dehalogenase

Zoom: 

Abstract:

Machine learning (ML) can guide the design of functional enzyme variants by elucidating sequence–activity relationships. Here, we apply a maximum-entropy generative model [1] to design single- and double-point mutants of a haloalkane dehalogenase and experimentally validate improved catalytic activity. Haloalkane dehalogenases play a crucial role in detoxifying biological media by converting toxic haloalkanes into harmless alcohols. Their enzymatic mechanism consists of two chemical steps: an SN2 nucleophilic attack followed by hydrolysis [2]. Kinetic experiments, including stopped-flow analysis, identified mutants that enhance catalytic activity either in the SN2 step or in the overall reaction, demonstrating the effectiveness of ML-driven enzyme engineering. These results are complemented by theoretical analyses using Empirical Valence Bond (EVB) and metadynamics methods, which reveal the structural, energetic, and electrostatic factors underlying activity changes in both native and engineered dehalogenase variants [3,4]. Overall, our study shows that ML-guided design, validated by both experimental kinetics and computational analysis, can effectively identify mutations that enhance enzymatic catalytic efficiency and provide mechanistic insight; however, larger mutational spaces still need to be explored.

[1]. W. Xie, M. Asadi, and A. Warshel PNAS 2022, 119, e2122355119, DOI

[2]. K. H. G. Verschueren et al Nature 1993, 363, 693–698, DOI

[3]. N. Gelfand et al J. Am. Chem. Soc. 2025, 147, 2747−2755, DOI

[4]. N. Gelfand, A. Warshel ACS Catal. 2025, 15, 13657−13666, DOI

Seminar Organizer: Prof. Ilia Kaminker