Computational Approaches for Drug-Protein Interaction Analysis in Cancer: Machine Learning and Structural Bioinformatics Perspectives

Abstract

Accurate drug-protein interaction is critical towards targeted cancer therapy and precision medicine. Conventional experimental methods for DPI identification are time-consuming and costly, necessitating computation approaches instead. Proposed in the study are: PCA-Cosine Similarity, an approach for DPI analysis, and CNN-Xception, a model for MRI-based brain tumor classification.The PCA-Cosine Similarity method uses PCA tools in performing dimension reduction of features while preserving predictive accuracy, rendering large-scale drug discovery more efficient. The CNN-Xception model blends CNN and Xception architecture via depthwise separable convolutions to offer improved tumor classification.

Experimental results show that PCA plus Cosine Similarity achieved 95.18% accuracy, outperforming raw similarity calculations very effectively and yet optimizing computational complexity. At the same time, the CNN-Xception model scored an impressive 100% across 656 test samples while differentiating between Glioma, Meningioma, Pituitary tumors, and non-tumor cases with great ease. From the comparative analysis, it is undoubted that deep-learning and similarity-based models work hand in hand, outperforming conventional methods in DPI prediction and brain tumor classification with a high degree of efficiency.Future enhancements will continue with molecular docking validation (AutoDock Vina) to tune DPI predictions, along with deep learning architecture integration to further such predictions. These developments will go a long way toward ushering in robust computational models for cancer drug discovery and precision oncology.