A Novel Ensemble Learning-Based Computational Method to Predict Protein-Protein Interactions from Protein Primary...

Area of Science:

• Computational biology
• Bioinformatics
• Proteomics

Background:

• Protein-protein interactions (PPIs) are fundamental to cellular functions but experimental detection methods are often time-consuming and error-prone.
• There is a significant need for accurate computational tools to supplement experimental PPI detection.
• Existing methods face challenges with high false negative rates, necessitating novel predictive approaches.

Purpose of the Study:

• To develop and validate a novel sequence-based computational model for predicting protein-protein interactions (PPIs).
• To combine Discrete Hilbert transform (DHT) and Rotation Forest (RoF) for enhanced PPI prediction accuracy.
• To evaluate the model's performance across different species and compare it with existing classifiers.

Main Methods:

• Utilized Position-Specific Scoring Matrices (PSSM) to encode protein evolutionary information from amino acid sequences.
• Constructed a 400-dimensional DHT descriptor for each protein pair to represent interaction features.
• Employed Rotation Forest (RoF) as a classifier to predict potential PPIs based on the generated descriptors.

Main Results:

• Achieved high prediction accuracies: 91.93% for Yeast, 96.35% for Human, and 94.24% for Oryza sativa PPI datasets.
• Demonstrated excellent predictive capacity on cross-species PPI datasets.
• Outperformed other classifiers like Support Vector Machine (SVM), Random Forest (RF), K-nearest Neighbor (KNN), and AdaBoost in comparative analyses.

Conclusions:

• The proposed DHT and RoF combined model is a highly accurate and feasible approach for predicting protein-protein interactions.
• This computational method serves as a valuable supplemental tool for large-scale proteomics analysis.
• The model's effectiveness across different species highlights its potential for broad application in biological research.