MESM: integrating multi-source data for high-accuracy protein-protein interactions prediction through multimodal...

Area of Science:

• Computational Biology
• Bioinformatics
• Machine Learning

Background:

• Protein-protein interactions (PPIs) are fundamental to numerous biological processes.
• Current prediction methods often lack comprehensive feature extraction, limiting their effectiveness.
• There is a need for advanced models to capture richer interaction information for improved PPI prediction.

Purpose of the Study:

• To develop a novel deep learning method, MESM, for enhanced protein-protein interaction (PPI) prediction.
• To overcome limitations of existing methods by integrating multimodal protein data.
• To improve the accuracy and robustness of PPI prediction models.

Main Methods:

• MESM employs multimodal representation extraction using Sequence Variational Autoencoder (SVAE), Variational Graph Autoencoder (VGAE), and PointNet Autoencoder (PAE).
• Fusion Autoencoder (FAE) integrates these multimodal features for balanced protein representations.
• GraphGPS, Graph Attention Network (GAT), Graph Convolutional Network (GCN), and SubgraphGCN are utilized to learn from PPI network structure and capture interaction details.

Main Results:

• MESM achieved significant performance improvements on benchmark datasets (SHS27k, SHS148k, SYS30k, SYS60k) from the STRING database.
• The model demonstrated superior accuracy in predicting protein-protein interactions compared to state-of-the-art methods.
• Integration of multimodal features and advanced graph learning modules contributed to enhanced prediction capabilities.

Conclusions:

• MESM represents a significant advancement in deep learning-based PPI prediction.
• The proposed method effectively integrates diverse protein data and network structures for improved accuracy.
• Experimental results validate MESM's potential for advancing biological research through accurate PPI identification.