EnzML: multi-label prediction of enzyme classes using InterPro signatures

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Manual annotation of enzymatic functions struggles to keep pace with rapid genome sequencing.
• There is a need for automated methods to predict protein enzymatic functions.

Purpose of the Study:

• To explore the capacity of InterPro sequence signatures for automated enzymatic function prediction.
• To develop and evaluate EnzML, a multi-label classification method for predicting enzymatic functions.

Main Methods:

• Utilized InterPro sequence signatures as features for machine learning.
• Developed EnzML, a multi-label classification approach.
• Evaluated EnzML on a dataset of 300,747 proteins with curated Enzyme Commission (EC) annotations.
• Tested EnzML on eight complete proteomes.

Main Results:

• EnzML achieved over 98% subset accuracy on the standard dataset.
• EnzML demonstrated 87-97% subset accuracy in reannotating eight diverse proteomes.
• Prediction accuracy remained robust even after reducing dataset redundancy using UniRef clusters.

Conclusions:

• InterPro signatures provide a powerful and compact feature space for enzymatic function prediction.
• The EnzML approach enables efficient multi-label machine learning for enzymatic function annotation.
• The method trains rapidly, making it feasible for large-scale genomic analysis.