Enhancing Enzyme Commission Number Prediction With Contrastive Learning and Agent Attention

Area of Science:

• Biochemistry
• Bioinformatics
• Computational Biology

Background:

• Accurate enzyme function prediction is vital for understanding diseases and drug target identification.
• Current enzyme commission (EC) number prediction methods face limitations in database coverage and sequence information mining, impacting annotation efficiency and precision.

Purpose of the Study:

• To introduce ProteEC-CLA, a novel model for predicting enzyme commission (EC) numbers using contrastive learning and agent attention.
• To enhance the accuracy and efficiency of enzyme function annotation through advanced sequence feature extraction and analysis.

Main Methods:

• ProteEC-CLA employs contrastive learning to create positive and negative sample pairs, improving sequence feature extraction and unlabeled data utilization.
• Integration of the ESM2 pre-trained protein language model generates informative sequence embeddings for functional correlation analysis.
• The Agent Attention mechanism is incorporated to enhance the capture of both local details and global features within enzyme sequences.

Main Results:

• ProteEC-CLA demonstrated exceptional performance on two independent datasets.
• Achieved 98.92% accuracy at the EC4 level on a standard dataset.
• On a challenging clustered split dataset, ProteEC-CLA attained 93.34% accuracy and a 94.72% F1-score.

Conclusions:

• ProteEC-CLA accurately predicts EC numbers up to the fourth level using only enzyme sequences as input.
• The model significantly enhances the efficiency and precision of enzyme functional annotation.
• ProteEC-CLA serves as a highly effective tool for enzymology research and applications.