SCREEN: A Graph-based Contrastive Learning Tool to Infer Catalytic Residues and Assess Enzyme Mutations

Area of Science:

• Biochemistry
• Bioinformatics
• Computational Biology

Background:

• Accurate identification of enzyme catalytic residues is crucial for understanding biological processes.
• The growing number of protein sequences requires automated computational tools for catalytic residue prediction.

Purpose of the Study:

• To introduce SCREEN, a novel graph neural network for high-throughput prediction of catalytic residues.
• To integrate enzyme functional and structural information for improved prediction accuracy.

Main Methods:

• SCREEN constructs residue representations based on spatial arrangements.
• Enzyme function priors are incorporated using contrastive learning.
• A graph neural network architecture is employed for prediction.

Main Results:

• SCREEN outperforms existing catalytic residue predictors.
• Accurate predictions are achieved even with inferred enzyme structures.
• The tool demonstrates good generalization to dissimilar enzymes.
• Predicted residues share characteristics with native catalytic residues.
• SCREEN can distinguish mutation-tolerant residues from those causing functional loss.

Conclusions:

• SCREEN is an effective tool for predicting catalytic residues, integrating structural and functional data.
• The predictions can potentially infer disease-associated mutations.
• SCREEN offers a valuable computational approach for enzyme research.