HECNet: a hierarchical approach to enzyme function classification using a Siamese Triplet Network.

Area of Science:

• Computational biology
• Enzymology
• Bioinformatics

Background:

• Enzyme function prediction is critical for biological research and industrial processes.
• Experimental determination of enzyme function is costly and time-consuming.
• Accurate enzyme function classification aids in disease research and commercial applications.

Purpose of the Study:

• To develop a novel computational approach for predicting enzyme function.
• To accurately classify enzyme function up to the fourth level of the Enzyme Commission (EC) Number.
• To address limitations in existing methods for detailed enzyme function prediction.

Main Methods:

• Utilized innovative deep learning techniques.
• Implemented an efficient hierarchical classification scheme.
• Trained and validated the model on a dataset of 11,353 enzymes across 402 classes.

Main Results:

• Achieved 91.2% hierarchical accuracy and 81.9% Macro-F1 score at the 4th EC Number level.
• Demonstrated success in predicting the function of enzyme isoforms.
• The method is broadly applicable for genome-wide enzyme function prediction.

Conclusions:

• The proposed deep learning approach accurately predicts enzyme function at a highly specific level.
• This methodology facilitates automated annotation of enzyme databases.
• Enables the identification of novel enzymes for commercial applications and research acceleration.