ProteinF3S: boosting enzyme function prediction by fusing protein sequence, structure, and surface.

Area of Science:

• Biochemistry and structural biology
• Computational biology and bioinformatics

Background:

• Proteins can be represented by sequence, structure, and surface data, each offering unique insights but with limitations.
• Integrating complementary information from these diverse data forms holds significant potential for advancing protein function prediction.

Purpose of the Study:

• To develop an effective framework, ProteinF3S, for enzyme function prediction by fusing information from protein sequence, structure, and surface data.
• To enhance feature representation through a multi-scale bidirectional fusion strategy between protein structure and surface representations.

Main Methods:

• Developed ProteinF3S, a framework integrating protein sequence, structure, and surface data for enzyme function prediction.
• Implemented a multi-scale bidirectional fusion strategy involving hierarchical features from surface and structure encoders.
• Achieved further feature fusion by concatenating sequence features with combined structure-surface features.

Main Results:

• ProteinF3S achieved new state-of-the-art performance on enzyme reaction classification and enzyme commission number prediction tasks.
• Demonstrated the effectiveness of fusing complementary information from sequence, structure, and surface data for improved enzyme function prediction.
• The multi-scale bidirectional fusion strategy yielded more distinctive and informative features.

Conclusions:

• The fusion of diverse protein data representations (sequence, structure, surface) is a highly effective strategy for enzyme function prediction.
• ProteinF3S provides a robust framework for leveraging multi-modal protein data, advancing the accuracy of enzyme function prediction.
• Future work can explore further refinements of fusion strategies and their application to other protein-related tasks.