Machine Learning Identifies Chemical Characteristics That Promote Enzyme Catalysis

Area of Science:

• Enzyme catalysis and engineering
• Computational biophysics
• Biochemistry

Background:

• Enzyme structure-dynamics-function relationships are not fully understood.
• Current enzyme engineering capabilities do not meet industrial demands.

Purpose of the Study:

• Investigate structural and dynamic factors driving catalysis in ketol-acid reductoisomerase (KARI).
• Identify enzyme-substrate complex conformations that enhance reactivity.
• Develop predictive models for enzyme reaction kinetics.

Main Methods:

• Computational statistical mechanics using transition interface sampling.
• Machine learning for feature selection and predictive modeling.
• Analysis of conformational and dynamic descriptors.

Main Results:

• Conformational descriptors alone predict reactivity with >85% accuracy (90% AUC).
• Key descriptors include substrate conformation, bond polarization, and metal coordination.
• Constraining trajectories to specific conformational regions increases computed reactivity significantly.

Conclusions:

• Reactivity-promoting regions exist within enzyme-substrate conformational space.
• Methodology developed for identifying and validating these regions.
• Enzyme re-engineering to populate these regions may yield substantial rate enhancements.