GTB-PPI: Predict Protein-protein Interactions Based on L1-regularized Logistic Regression and Gradient Tree Boosting

Area of Science:

• Proteomics
• Bioinformatics
• Computational Biology

Background:

• Protein-protein interactions (PPIs) are fundamental to understanding biological processes, disease mechanisms, and drug development.
• The growing volume of PPI data and advancements in machine learning have spurred research into computational prediction methods.

Purpose of the Study:

• To develop a novel and accurate prediction pipeline for protein-protein interactions (PPIs).
• To leverage gradient tree boosting (GTB) for enhanced PPI prediction accuracy.

Main Methods:

• Feature extraction through fusion of pseudo amino acid composition (PseAAC), pseudo position-specific scoring matrix (PsePSSM), reduced sequence and index-vectors (RSIV), and autocorrelation descriptor (AD).
• Feature subset selection using L1-regularized logistic regression (L1-RLR) to minimize redundancy and noise.
• Construction of the Gradient Tree Boosting for Protein-Protein Interactions (GTB-PPI) prediction model.

Main Results:

• GTB-PPI achieved high accuracies of 95.15% on Saccharomyces cerevisiae and 90.47% on Helicobacter pylori datasets via five-fold cross-validation.
• The model demonstrated effectiveness on independent test datasets across multiple species (Caenorhabditis elegans, Escherichia coli, Homo sapiens, Mus musculus).
• Successful application to predict specific PPI networks, including CD9 and Wnt-related signaling pathways, confirming significant accuracy improvements.

Conclusions:

• The proposed GTB-PPI pipeline offers a robust and accurate method for predicting protein-protein interactions.
• This approach has broad applicability in proteomics research, disease pathogenesis studies, and drug design.
• The developed tool and datasets are publicly available to facilitate further research in PPI prediction.