GloEC: a hierarchical-aware global model for predicting enzyme function.

Area of Science:

• Biotechnology
• Bioinformatics
• Computational Biology

Background:

• Enzyme function annotation is crucial for industrial biotechnology and understanding pathologies.
• Existing computational methods struggle to model the hierarchical nature of enzyme labels and their inter-level interactions.
• Accurate enzyme function prediction requires a global perspective on enzyme label dependencies.

Purpose of the Study:

• To develop a novel computational model for predicting enzyme function that addresses limitations of existing methods.
• To effectively model the hierarchical structure and interdependencies of enzyme labels.
• To improve the accuracy and scope of enzyme function prediction, including isoenzyme identification.

Main Methods:

• Formulated enzyme label hierarchy as a directed enzyme graph.
• Proposed a hierarchy-based Graph Convolutional Network (GCN) encoder for global dependency modeling.
• Developed an end-to-end hierarchical-aware global model named GloEC.
• Implemented bidirectional computation in the hierarchy-GCN encoder for bottom-up and top-down analysis.

Main Results:

• GloEC achieved superior predictive performance compared to existing methods across three benchmark datasets.
• The model demonstrated effectiveness in predicting the functions of isoenzymes through case studies.
• Hierarchical-aware embeddings and deductive fusion of features enhanced prediction accuracy.

Conclusions:

• GloEC offers a significant advancement in enzyme function prediction by incorporating hierarchical information.
• The bidirectional computation approach in the hierarchy-GCN encoder captures novel enzyme label correlations.
• The model provides a robust and accurate solution for enzyme function annotation in computational biology.