Machine learning-based prediction of enzyme substrate scope: Application to bacterial nitrilases

Area of Science:

• Biochemistry
• Computational Biology
• Enzymology

Background:

• Predicting enzyme substrate specificity from amino acid sequences is challenging, especially for closely related molecules.
• Current sequence- and structure-based methods often fail to predict specific substrate acceptance.

Purpose of the Study:

• To develop and evaluate a computational approach for accurately predicting the substrate scope of bacterial nitrilases.
• To integrate experimental data, structural modeling, and machine learning for enhanced enzyme substrate prediction.

Main Methods:

• Combined targeted experimental activity data with structural modeling and ligand docking.
• Utilized physicochemical properties of proteins and ligands.
• Applied and compared four machine learning models: logistic regression, random forest, gradient-boosted decision trees, and support vector machines.

Main Results:

• The developed approach achieved high predictive performance for bacterial nitrilase substrate scope.
• All four machine learning models demonstrated similar performance, with an average ROC of 0.9 and accuracy of approximately 82%.
• Random forest models showed some advantages in predictive capability.

Conclusions:

• The integrated approach effectively predicts substrate scope for bacterial nitrilases.
• This modular methodology can be adapted for predicting substrate specificity in other enzyme families.
• Machine learning combined with structural and physicochemical data offers a powerful tool for enzyme engineering and discovery.