Multi-DeepProtGraphGO: Integrating GCN on PPI Networks With Sequence-Driven Convolutional Bi-LSTM and Attention for...

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Protein function prediction is crucial for understanding biological processes and disease mechanisms.
• A significant gap exists between the rapid discovery of new proteins and the annotation of their functions.
• Existing methods often overlook valuable protein-protein interaction (PPI) network data and neighborhood information.

Purpose of the Study:

• To develop a novel multi-modal approach for enhanced protein function prediction.
• To leverage both PPI network topology and protein sequence information.
• To address limitations of current methods that rely solely on node2vec embeddings.

Main Methods:

• Utilized a Graph Convolutional Network (GCN) to analyze PPI network data and protein neighborhood relationships.
• Employed Multi-Head Self-Attention with a Convolutional Bi-LSTM on protein sequences.
• Integrated these approaches into a novel multi-modal framework named Multi-DeepProtGraphGO.
• Validated using benchmark datasets from Homo sapiens (String database for PPI, UniprotKB for sequences).

Main Results:

• The proposed Multi-DeepProtGraphGO method demonstrated significant improvements over the state-of-the-art.
• Achieved +18.28% Fmax score improvement for Biological Process (BP).
• Achieved +4.56% Fmax score improvement for Cellular Component (CC).
• Achieved +6.92% Fmax score improvement for Molecular Function (MF).

Conclusions:

• The novel multi-modal approach effectively integrates PPI network and protein sequence data for superior protein function prediction.
• Multi-DeepProtGraphGO outperforms existing methods, offering a more robust tool for bioinformatics research.
• This advancement aids in classifying proteins and understanding their roles in disease mechanisms.