DMMAFS: Protein Function Prediction Based on Multi-Modal Multi-Attention Fusion Features.

Area of Science:

• Computational Biology
• Bioinformatics
• Structural Biology

Background:

• Protein function prediction is crucial for biological research.
• Current methods often rely solely on protein sequence data, limiting accuracy.
• Existing multi-modal approaches may not fully leverage complementary information.

Purpose of the Study:

• To develop an advanced deep learning model for protein function prediction.
• To effectively integrate diverse protein data modalities, including sequence and structure.
• To overcome limitations of existing methods in feature fusion and information exploitation.

Main Methods:

• Proposing Multi-modal Multi-attention fusion Features (DMMAFS), a deep learning framework.
• Utilizing self-attention mechanisms to extract semantic information from protein sequences.
• Employing a S-C cross-modal cross-attention fusion network to integrate sequence and 3D structural information.

Main Results:

• DMMAFS effectively captures semantic information from protein sequences.
• The method successfully compensates for sequence-based predictions using 3D structural data.
• Experimental results show DMMAFS outperforms state-of-the-art methods in protein function prediction.

Conclusions:

• DMMAFS offers a novel and effective approach to protein function prediction.
• Integrating multi-modal data, particularly sequence and structure, enhances prediction accuracy.
• The proposed cross-attention fusion mechanism is key to exploiting complementary information between modalities.