StarFunc: Fusing Template-based and Deep Learning Approaches for Accurate Protein Function Prediction.

Area of Science:

• Computational biology
• Bioinformatics
• Structural biology

Background:

• Deep learning methods have advanced protein function prediction.
• Template information, often from sequence homology or protein-protein interactions, is crucial for current methods.
• Few methods utilize structural similarity for template identification, despite structure-function relationships.

Purpose of the Study:

• To develop StarFunc, a composite approach for protein function prediction.
• To integrate state-of-the-art deep learning with multiple template information sources.
• To evaluate StarFunc's performance against existing methods.

Main Methods:

• Developed StarFunc, a composite prediction approach.
• Integrated deep learning models with template data from sequence homology, protein-protein interactions, structural similarity, and protein domain families.
• Conducted large-scale benchmarking and blind testing using the 5th Critical Assessment of Function Annotation (CAFA5) dataset.

Main Results:

• StarFunc demonstrated superior performance in protein function prediction.
• The composite approach showed advantages over state-of-the-art deep learning methods.
• StarFunc outperformed conventional template-based predictors in CAFA5 evaluations.

Conclusions:

• Integrating diverse template information with deep learning enhances protein function prediction accuracy.
• StarFunc represents a significant advancement in computational approaches for predicting protein functions.
• The study highlights the importance of structural similarity in template identification for function prediction.