Improved enzyme functional annotation prediction using contrastive learning with structural inference.

Area of Science:

• Computational Biology
• Bioinformatics
• Deep Learning Applications

Background:

• Deep learning has advanced scientific research, with significant applications in predicting enzyme function.
• Current computational methods for enzyme function prediction often rely on either amino acid sequences or protein structures independently.
• There is a need to integrate diverse data modalities for more accurate enzyme function prediction.

Purpose of the Study:

• To develop a novel deep learning framework that combines amino acid sequence and protein contact map data for enzyme function prediction.
• To address the limitations of existing methods that focus on single data types.

Main Methods:

• Proposed a Contrastive Learning framework for Enzyme functional ANnotation prediction combined with protein amino acid sequences and Contact maps (CLEAN-Contact).
• Evaluated CLEAN-Contact against state-of-the-art enzyme function prediction models using multiple benchmark datasets.
• Applied the framework to predict novel enzyme functions in the Prochlorococcus marinus MED4 proteome.

Main Results:

• CLEAN-Contact demonstrated superior performance compared to existing state-of-the-art enzyme function prediction models.
• The framework successfully predicted previously unknown enzyme functions.
• Integration of sequence and contact map data significantly enhanced prediction accuracy.

Conclusions:

• The CLEAN-Contact framework represents a significant advancement in enzyme function prediction accuracy.
• Combining multiple data modalities, such as amino acid sequences and contact maps, is crucial for robust enzyme function prediction.
• This approach holds promise for uncovering novel enzymatic functions in uncharacterized proteomes.