Multi-View Collaboration Feature Fusion for Protein Function Prediction.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• High-throughput sequencing generates vast amounts of uncharacterized protein data.
• Experimental protein function validation is resource-intensive and slow.
• Existing Automatic Function Prediction (AFP) methods struggle with complex, multi-level protein functions and Gene Ontology (GO) label relationships.

Purpose of the Study:

• To develop an advanced framework for enhancing protein function prediction.
• To address limitations in current AFP methods regarding feature representation and hierarchical label dependencies.
• To improve the accuracy and generalization of predicting protein functions.

Main Methods:

• Proposed a Multi-View Collaboration Feature Fusion (MVCFF) framework.
• Utilized complementary features from multiple protein sequence perspectives.
• Developed a sequential feature extraction subnetwork for view-specific information capture (local patterns and long-range dependencies).
• Implemented a multi-view collaboration paradigm for interactive learning and synergistic information fusion.
• Introduced MVCFF+ by combining MVCFF with sequence-similarity-based methods via weighted fusion.

Main Results:

• MVCFF effectively leverages complementary sequence features for enhanced protein function prediction.
• The multi-view collaboration paradigm facilitates synergistic information fusion.
• MVCFF+ demonstrated superior performance by integrating sequence-similarity methods.
• Extensive experiments confirmed substantial improvements in prediction performance compared to existing methods.

Conclusions:

• The MVCFF framework offers a powerful approach to overcome limitations in current AFP.
• Integrating diverse sequence features and employing collaborative learning significantly boosts prediction accuracy.
• The proposed methods represent a significant advancement in computational protein function prediction.