Drug Recommendation System for Precision Medicine using Graph Neural Network with Hierarchical Proximal Policy Optimization

In recent years, precision medicine aims to enhance cancer treatment based on the molecular profile of individual tumors but identifying the most effective drug for a specific cancer type remains a significant challenge. However existing models such as Deep Neural Networks (DNN), Recurrent Neural Networks (RNN) and other existing model struggle to capture the complex, heterogeneous interactions among drugs, genes, and cell lines, and often lack interpretability and biological reasoning. To resolve these issues, a novel framework integrating Graph Neural Networks (GNN) with Hierarchical Proximal Policy Optimization (HPPO) which is named as GNN-HPPO is proposed. Initially, the data is collected from the Genomics of Drug Sensitivity in Cancer (GDSC) and Cancer Cell Line Encyclopedia (CCLE) datasets which comprises of drug response data and molecular profiles of cancer cell lines. Initially, in the preprocessing stage, gene expression and drug features are log-transformed to determine drug sensitivity, then normalized using min-max normalization. Further, a heterogeneous graph is constructed to model interactions between drugs, genes, and cell lines, and a GNN is applied to learn relational embeddings. Furthermore, for drug recommendation, HPPO is incorporated where a high-level policy selects drug classes and a low-level policy recommends specific drugs.