A novel one-class SVM based negative data sampling method for reconstructing proteome-wide HTLV-human protein...

Area of Science:

• Computational Biology
• Bioinformatics
• Systems Biology

Background:

• Protein-protein interaction (PPI) prediction is crucial for understanding biological processes.
• Current methods often struggle with accurate negative data sampling and minimizing false positives.
• Predicting interactions between retroviruses like HTLV and host proteins is vital for understanding pathogenesis.

Purpose of the Study:

• To develop a novel negative data sampling method for PPI prediction.
• To improve the accuracy and reduce false positives in computational PPI prediction.
• To predict proteome-wide interactions between HTLV and Homo sapiens.

Main Methods:

• Utilized one-class Support Vector Machine (SVM) for reliable negative data sampling.
• Employed two-class SVM for proteome-wide prediction and model selection feedback.
• Applied gene ontology-based clustering to analyze predicted PPI networks.

Main Results:

• One-class SVM demonstrated superior performance for negative data sampling compared to two-class predictors.
• Predictive feedback-constrained model selection effectively reduced false positive predictions.
• Several predicted PPIs were validated by recent literature.
• Gene ontology analysis provided insights into HTLV pathogenesis.

Conclusions:

• The proposed one-class SVM-based negative sampling method enhances PPI prediction accuracy.
• Rational model selection guided by predictive feedback minimizes false positives.
• The predicted HTLV-human PPI network offers valuable insights into viral pathogenesis.