TOP-BIOCom: A Feature Fusion-based Prediction of Protein Complexes from PPI Networks

Protein-Protein Interactions (PPI) are crucial for cellular functions. Computational prediction of protein complexes from PPI networks is essential, yet traditional methods relying solely on network topology often lack biological features. Integrating topological and biological features can enhance prediction accuracy.

We proposed TOP-BIOCom, a machine learning-based approach that integrates feature fusion of novel topological, structural, and sequence-based features with the Embedding Lookup technique. The benchmark dataset was CYC2008, while the PPI network datasets were DIP and BioGrid. The performance evaluation measures precision, recall, and F-1 score were carried out to assess the efficiency of the TOP-BIOcom model and compared with the reported models.

Our result with a novel feature fusion approach, demonstrated that the BioGrid PPI network dataset with Random Forest yielded an accuracy of 0.99, precision of 0.96, recall of 0.97, and an F1-score of 0.96. The model's validation accuracy was 0.99 and completed the task in 3.85 seconds. DIP dataset with LightGBM model achieved an accuracy of 0.95, with a precision of 0.88, a recall of 0.91, and an F1-score of 0.89. The validation accuracy matched the accuracy at 0.95.

These results highlight the robustness of the proposed TOP-BIOcom model in predicting protein complexes from PPI networks with higher accuracy and faster execution. The proposed approach demonstrates superiority over existing methods, showing its effectiveness across different datasets and machine learning models.

These findings suggest that integrating topological and biological features can provide a holistic view of protein complexes enhancing prediction accuracy and aiding in drug discovery and understanding cellular mechanisms.