MMSMAPlus: a multi-view multi-scale multi-attention embedding model for protein function prediction

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Protein function prediction is crucial for understanding biological processes.
• Existing intelligent methods utilize diverse protein data (sequence, structure, network).
• Exploiting multi-view features from protein sequences remains a challenge.

Purpose of the Study:

• To propose a novel deep neural model for enhanced protein function prediction.
• To effectively integrate multi-view, multi-scale, and multi-attention mechanisms for sequence analysis.
• To improve the accuracy and comprehensiveness of protein function prediction.

Main Methods:

• Developed a multi-view, multi-scale, multi-attention deep neural model (MMSMA).
• Extracted multi-view features: one-hot encoding, evolutionary, deep semantic, and physiochemical properties.
• Integrated homology-based prediction into an extended model (MMSMAPlus).

Main Results:

• MMSMA captures multi-scale local patterns and long-range dependencies in protein sequences.
• The MMSMAPlus model demonstrated superior performance compared to state-of-the-art methods.
• Comprehensive decision-making was achieved through a multi-view adaptive mechanism.

Conclusions:

• The proposed MMSMAPlus model offers a significant advancement in protein function prediction.
• Multi-view deep learning effectively leverages diverse protein sequence information.
• The model provides a robust framework for future bioinformatics research.