Novel Protein-Protein Interaction Prediction with Updated Prism Refraction Search with Compression-based Graph Convoluted Radial Basis Function Model

Abstract

Protein-Protein Interactions (PPIs) play a pivotal role in understanding biological phenomena, but the existing approaches for their detection are often expensive, prone to false positives, and computationally intensive. Moreover, the existing computational models are faced with scalability and interpretability problems, which impede their widespread use in biological discovery and drug design. To overcome these hurdles, this research proposes the Compression-Based Graph Convoluted Radial Basis Function (CCRBF) Framework, which combines the use of graph convolutional networks and radial basis functions to represent the complex, non-linear relationships between proteins. This improves the prediction accuracy by learning the complex patterns of interactions. In addition, the Updated Prism Refraction Search Optimizer (UPRSO) is also used to dynamically modify the model parameters during the optimization process. Moreover, the model uses SHapley Additive exPlanations (SHAP), an explainable AI tool that offers interpretability in the decision-making process. SHAP assists in understanding the role of individual protein attributes in PPI prediction, which enhances the biological validity and authenticity of the model. The CCRBF model performs outstandingly with 99.95% accuracy, 99.80% precision, 99.98% sensitivity, and 99.87% F1-score, which is an efficient approach for large-scale PPI prediction, thereby contributing to biological research.