Using the Relevance Vector Machine Model Combined with Local Phase Quantization to Predict Protein-Protein...

Area of Science:

• Computational biology
• Bioinformatics
• Proteomics

Background:

• Protein-protein interactions (PPIs) are crucial for cellular functions.
• Accurate prediction of PPIs from protein sequences is essential for understanding biological processes.
• Existing methods face challenges in accuracy and noise reduction.

Purpose of the Study:

• To introduce a novel computational method, RVM-LPQ, for predicting PPIs.
• To enhance the accuracy and robustness of PPI prediction using advanced feature representation and classification techniques.

Main Methods:

• Utilizing Local Phase Quantization (LPQ) for protein sequence feature representation on Position Specific Scoring Matrices (PSSM).
• Applying Principal Component Analysis (PCA) to reduce noise and improve feature representation.
• Employing a Relevance Vector Machine (RVM) as the core classification model.

Main Results:

• Achieved high accuracies of 92.65% on Yeast datasets and 97.62% on Human datasets.
• Demonstrated superior performance compared to previous methods and state-of-the-art Support Vector Machine (SVM) classifiers.
• Validated the method's efficiency and simplicity through rigorous 5-fold cross-validation experiments.

Conclusions:

• The RVM-LPQ method offers a significant advancement in predicting PPIs from protein sequences.
• The proposed approach is efficient, simple, and highly accurate.
• RVM-LPQ can serve as an effective decision support tool for future proteomics research.