A Sub-Sequence Based Approach to Protein Function Prediction via Multi-Attention Based Multi-Aspect Network.

Area of Science:

• Bioinformatics
• Computational Biology
• Protein Science

Background:

• Protein function prediction is crucial for understanding biological processes.
• Classifying protein sub-sequences is challenging due to limited knowledge of sub-sequence functions.
• Existing methods often struggle to efficiently utilize parent sequence information for sub-sequence classification.

Purpose of the Study:

• To develop a novel "multi-aspect" paradigm for efficient protein sub-sequence classification.
• To propose the "Multi-Attention Based Multi-Aspect Network" for sub-sequence classification.
• To introduce the Global-ProtEnc method for protein sequence encoding and function prediction.

Main Methods:

• Developed a "multi-aspect" paradigm incorporating multi-label and label-relevance scoring.
• Proposed the "Multi-Attention Based Multi-Aspect Network" utilizing word-level multi-attention for sub-sequence processing.
• Introduced Global-ProtEnc, a sub-sequence-based encoding approach, and its ensemble version, Global-ProtEnc-Plus.

Main Results:

• Global-ProtEnc and Global-ProtEnc-Plus demonstrated outstanding performance on the CAFA3 benchmark dataset.
• Global-ProtEnc-Plus showed significant improvements over the state-of-the-art DeepGOPlus.
• Specific improvements include +6.50% Fmax for biological process and +1.90% for cellular component prediction.

Conclusions:

• The proposed multi-aspect paradigm and Global-ProtEnc methods enhance protein sub-sequence classification and function prediction.
• Global-ProtEnc-Plus represents a significant advancement in protein function prediction accuracy.
• These methods offer efficient and effective tools for analyzing protein sequences and their functions.