Predicting Protein-Protein Interactions Using Vectorized Sequence Descriptors and Machine Learning Classifiers

Random Forest using 120 Auto-Covariance, 686 Conjoint Triad, and 1,820 Local Descriptor features predicts protein-protein interactions on the Human-PPI test set with 99.51% accuracy and AUROC 0.9990.

• Why it matters: Experimental PPI mapping is costly and slow, creating a need for computational methods that can reliably generalize while remaining interpretable for interactome screening.
• What they did: The authors trained interpretable machine-learning classifiers (SVM, Random Forest, XGBoost) on a gold-standard sequence-based PPI benchmark using stratified five-fold cross-validation, then evaluated on an independent Human-PPI test set.
• The result: Random Forest achieved 99.51% accuracy, 99.72% precision, 99.29% recall, 99.50% F1, AUPRC 0.9989, and feature-importance highlighted short-range physicochemical correlations and specific triad patterns.