Improving protein-protein interaction prediction using protein language model and protein network features.

Area of Science:

• Bioinformatics
• Computational Biology
• Systems Biology

Background:

• Protein-protein interactions (PPIs) are crucial for biological processes.
• Experimental PPI identification is costly and time-consuming.
• Computational methods are needed to accelerate PPI prediction.

Purpose of the Study:

• To develop a novel hybrid computational method (KSGPPI) for improved protein-protein interaction prediction.
• To leverage protein sequences and interaction networks for enhanced prediction accuracy.
• To provide a more efficient alternative to experimental PPI identification.

Main Methods:

• Utilized ESM-2 (a protein language model) and CKSAAP with a 2D CNN for sequence-based feature extraction.
• Employed NW-align and Node2vec on the STRING database for network-based feature extraction.
• Fused features from both modules and used a multilayer perceptron for PPI prediction.

Main Results:

• Achieved an average prediction accuracy of 88.96% via five-fold cross-validation.
• Demonstrated a significantly higher Matthews correlation coefficient (0.781) compared to state-of-the-art methods.
• The KSGPPI tool is available as a standalone package.

Conclusions:

• The KSGPPI method effectively predicts protein-protein interactions by integrating sequence and network information.
• This hybrid approach offers a promising computational strategy for advancing PPI research.
• KSGPPI provides a valuable tool for biological mechanism studies and drug discovery.