EZpred: improving deep learning-based enzyme function prediction using unlabeled sequence homologs.

Area of Science:

• Bioinformatics
• Computational Biology
• Machine Learning

Background:

• Deep learning models excel at protein structure prediction using sequence homolog features.
• However, sequence homolog features are underutilized in protein function prediction.
• Existing methods struggle to leverage unlabeled sequence homologs for function prediction.

Purpose of the Study:

• To develop the first deep learning model, EZpred, for protein function prediction utilizing unlabeled sequence homologs.
• To enhance the accuracy of Enzyme Commission (EC) number prediction.

Main Methods:

• EZpred identifies sequence homologs using MMseqs2.
• Sequence features are extracted via the ESMC protein language model.
• A deep learning model predicts EC numbers using these features.

Main Results:

• EZpred achieved a 4% higher F1-score compared to models not using sequence homologs.
• EZpred outperformed state-of-the-art EC number prediction models by at least 10%.
• The study demonstrated the significant impact of sequence homologs on enzyme function prediction accuracy.

Conclusions:

• Unlabeled sequence homologs are valuable for deep learning-based protein function prediction.
• EZpred represents a significant advancement in predicting Enzyme Commission numbers.
• The findings highlight a new direction for improving protein function prediction models.