Prot2GO: Predicting GO Annotations From Protein Sequences and Interactions

Area of Science:

• Bioinformatics
• Computational Biology
• Molecular Biology

Background:

• Proteins are fundamental to life, and understanding their functions is vital for drug discovery, disease treatment, and vaccine development.
• Deep learning models have advanced protein function prediction, but often rely solely on protein sequences, limiting multi-source data integration.
• Existing methods struggle to effectively annotate protein functions by combining diverse data types.

Purpose of the Study:

• To develop an advanced deep learning model, Prot2GO, for accurate protein function prediction.
• To integrate both protein sequence data and protein-protein interaction (PPI) network data for enhanced functional annotation.
• To improve the predictive power of computational models for protein functions.

Main Methods:

• Prot2GO utilizes an improved biased random walk algorithm for PPI network feature extraction.
• Convolutional Neural Networks (CNNs) capture local sequence features, while Recurrent Neural Networks (RNNs) identify long-range residue associations.
• An attention mechanism is incorporated to identify critical protein motifs and structural domains.

Main Results:

• Prot2GO demonstrates superior performance compared to existing methods across multiple evaluation metrics.
• The model effectively integrates multi-source data, including sequence and PPI network information.
• Feature extraction from PPI networks and sequence data is significantly enhanced by the proposed methods.

Conclusions:

• Prot2GO represents a significant advancement in protein function prediction by effectively integrating diverse data sources.
• The model's ability to leverage both sequence and network information offers a more comprehensive approach to functional annotation.
• This work provides a powerful new tool for bioinformatics research, with implications for medicine and biotechnology.