RC-GNN: A predictive model of enzyme-reaction pairs

Area of Science:

• Biochemistry
• Bioinformatics
• Computational Biology

Background:

• Uncharacterized enzyme functions offer significant potential for developing novel therapeutics, sustainable materials, and understanding metabolic evolution.
• The generation of novel enzymes and reactions (de novo) by computational tools presents a vast, yet challenging, area for exploration.
• High-throughput screening for enzymatic activity is technically complex, necessitating predictive models for in silico pre-screening.

Purpose of the Study:

• To develop and evaluate a predictive model for identifying enzyme-catalyzed reactions, particularly for de novo enzyme-reaction pairs.
• To assess the generalization capabilities of the model on unseen enzymes and reactions.

Main Methods:

• Development of the Reaction-Center Graph Neural Network (RC-GNN) model.
• Representation of enzymes by amino acid sequences and reactions by their reactants and products.
• In silico evaluation of RC-GNN's predictive accuracy on de novo enzyme-reaction pairs under varying similarity conditions.

Main Results:

• RC-GNN demonstrated strong predictive performance, achieving 78.0% accuracy for novel reactions and 94.8% accuracy for novel enzymes when similarity was controlled.
• The model showed significant generalization capabilities, accurately predicting catalysis for enzymes and reactions distinct from the training data.
• Performance was robust even under challenging conditions with high dissimilarity between training and testing data.

Conclusions:

• RC-GNN is a powerful tool for predicting enzymatic activity on de novo reactions, facilitating the exploration of uncharacterized enzyme functions.
• The model's ability to generalize makes it valuable for metabolic engineers and evolutionary biologists seeking to understand and engineer enzymatic processes.
• This work advances in silico screening methods, paving the way for accelerated discovery in enzyme engineering and synthetic biology.