Answers are grounded in this paper and cite their sources.
The study of Host-Pathogen Protein-Protein Interactions (HP-PPIs) is critical for understanding mechanisms of infectious disease and for the development of therapeutic interventions. Computational prediction of HP-PPIs has become essential due to the limitations of experimental approaches, which are labour-intensive and expensive. In this tutorial, we present a step-by-step machine learning pipeline for HP-PPI prediction implemented in Google Colab. The workflow consists of six stages implemented in six Google Colab notebooks: dataset preparation, data preprocessing, feature extraction, machine learning model training, hyperparameter tuning and validation and model deployment. Six widely used classifiers, including LightGBM, Multi-layer Perceptron, Logistic Regression, Support Vector Machine, Random Forest, and XGBoost, were evaluated using balanced and imbalanced datasets from human-bacteria protein interactions. Amino acid composition was evaluated as a standalone feature descriptor, while pseudo amino acid and conjoint triad descriptors were concatenated into a second feature set. Among the classifiers, LightGBM achieved the highest mean cross-validation performance Matthews correlation coefficient (MCC) and was subsequently selected for evaluation on an independent test set. Leave-One-Pathogen-Out validation was additionally performed to assess the ability of the selected model to generalise to unseen bacterial pathogens. This tutorial provides a reproducible and adaptable framework for HP-PPI prediction that can be extended to other pathogen systems. It is beginner-friendly and provides a starting point for researchers with limited computational backgrounds exploring the research domain of infectious disease biology.
No takes yet. Share an insight, caveat, or question.
Emmanuel et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: