This is the authors' abstract. We don't add key points for this paper.
Background Protein–protein interactions are essential for understanding cellular processes and molecular mechanisms in living organisms. In Escherichia coli , accurate identification of PPIs is crucial for elucidating functional interaction networks and disease-associated mechanisms. Objectives This study aims to develop and evaluate a machine learning framework for predicting PPIs using Amino Acid Composition (AAC) features and to systematically compare classifier performance under different validation strategies. Methods Experimentally validated PPI data were obtained from the STRING database, and protein sequences were retrieved from UniProtKB. Balanced datasets were constructed using different positive-to-unknown sampling ratios to address class imbalance. Five supervised classifiers, including Decision Tree (DT), Random Forest (RF), Support Vector Machine (SVM), Extreme Gradient Boosting (XGBoost) and LightGBM, were trained and evaluated using 10-fold cross-validation as the primary evaluation strategy, complemented by independent test datasets for robust performance assessment. Results Cross-validation showed that ensemble models generally outperformed traditional classifiers in stability and generalization. RF showed a marked performance drop (accuracy ≈ 0.51, AUC ≈ 0.69), indicating overfitting and sensitivity to data partitioning. In contrast, XGBoost and LightGBM were more consistent, with accuracies of 0.66 and 0.65 and AUC values of 0.683 and 0.697. On independent test data, all models achieved moderate performance, with AUC values of 0.71 (RF), 0.70 (XGBoost), and 0.69 (LightGBM), highlighting the limitations of sequence-only features for PPI prediction. Conclusions AAC-based machine learning models provide a computationally efficient baseline for large-scale microbial PPI prediction. The study highlights the importance of rigorous validation and the limitations of sequence-derived features, suggesting that richer biological representations may improve future performance.
Answers are grounded in this paper and cite their sources.
No takes yet. Share an insight, caveat, or question.
Tasmia et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: