Is EC class predictable from reaction mechanism?

Area of Science:

• Enzyme kinetics and classification
• Cheminformatics
• Machine learning in biochemistry

Background:

• Investigating relationships between Enzyme Commission (EC) class, chemical reactions, and reaction mechanisms.
• Utilizing predictive models such as Support Vector Machine (SVM), Random Forest (RF), and k-Nearest Neighbours (kNN).
• Exploring different descriptor encoding strategies for reaction mechanisms and overall chemical reactions.

Purpose of the Study:

• To build and evaluate predictive models for enzyme classification based on reaction characteristics.
• To compare the effectiveness of different machine learning algorithms (SVM, RF, kNN) in predicting enzyme function.
• To assess the impact of descriptor types (mechanism vs. overall reaction) on prediction accuracy.

Main Methods:

• Development of predictive models using SVM, RF, and kNN.
• Encoding of reaction mechanisms and overall chemical reactions into distinct descriptor sets.
• Validation using both cross-validation and an independent external test set.

Main Results:

• Descriptor sets encoding overall chemical transformations outperformed those describing reaction mechanisms.
• SVM and RF models demonstrated comparable and strong performance; kNN was less successful.
• Oxidoreductases and hydrolases were well-predicted across descriptor types; isomerases were better predicted by overall reaction descriptors.

Conclusions:

• Enzyme reaction similarity does not guarantee similar mechanisms; EC number prediction is multifaceted.
• Enzyme classes like oxidoreductases, hydrolases, isomerases, and ligases possess clearer chemical signatures for prediction.
• Isomerases exhibit significant mechanistic diversity, challenging simple classification based on their shared isomer property.