An Improved Bagging of Machine Learning Algorithms to Predict Motif Structures From Protein-Protein Interaction Networks

Protein-protein interaction (PPI) networks play a vital role in the functional organization of cellular systems. Recurring subgraph patterns with these networks, known as motif structures, can be used to uncover mechanisms in complex biological processes. However, predicting motif structures from PPI networks remains challenging because of the complex and dynamic nature of interactions. Hence, this research focuses on predicting the topological structure of protein-protein interaction networks, acknowledging the computational challenges of exact motif discovery for diverse protein interactions. This study proposes a hybrid ensemble classification algorithm that leverages Penalized Matrix Decomposition (PMD) for feature extraction and a bagging approach using a Support Vector Machine (SVM), Alternative Decision Tree (ADT), and Extreme Learning Machine (ELM) classifiers. The proposed hybrid classification algorithm was evaluated against the ACC-Motif, Combi-Motif, Minimum Spanning Tree (MST), and Slider on benchmark datasets to demonstrate its statistical significance. On benchmark datasets, the proposed method achieved 92.3% accuracy, 91.5% precision, 89.8% recall, and an F1-score of 90.6%, outperforming existing algorithms such as ACC-Motif, Combi-Motif, and MST by 8-10%. This study assessed the practicality of similar topological subnetworks in protein-protein interaction networks, with Z-test calculations affirming superior outcomes. Overall, the proposed algorithm proved effective in predicting network structures, showing improved performance compared with other methods.