Enzyme specificity prediction using cross-attention graph neural networks.

Area of Science:

• Biochemistry
• Bioinformatics
• Computational Biology

Background:

• Enzyme function is governed by substrate specificity, determined by active site structure and reaction transition states.
• Millions of enzymes lack substrate specificity data, hindering research and applications.
• Existing prediction models have limitations in accuracy and scope.

Purpose of the Study:

• To develop a novel machine learning model for accurate enzyme substrate specificity prediction.
• To create a comprehensive database of enzyme-substrate interactions for model training.
• To overcome limitations of existing methods and improve biocatalytic understanding.

Main Methods:

• Developed EZSpecificity, a cross-attention-empowered SE(3)-equivariant graph neural network.
• Trained the model on a custom database of enzyme-substrate interactions at sequence and structural levels.
• Validated EZSpecificity against existing models using diverse enzyme and substrate datasets.

Main Results:

• EZSpecificity significantly outperformed state-of-the-art machine learning models.
• Experimental validation with eight halogenases and 78 substrates showed 91.7% accuracy in predicting reactive substrates.
• The model demonstrated superior performance on unknown enzyme and substrate databases and proof-of-concept protein families.

Conclusions:

• EZSpecificity provides a general and accurate method for predicting enzyme substrate specificity.
• The model has broad applications in fundamental and applied research in biology and medicine.
• This work advances the understanding and utilization of enzymes in biocatalysis.