Graph Attention Networks for Drug Combination Discovery: Targeting Pancreatic Cancer Genes with RAIN Protocol

Background Malignant neoplasm of the pancreas (MNP), a highly lethal illness with bleak outlook and few therapeutic avenues, entails numerous cellular transformations. These include irregular proliferation of ductal cells, activation of stellate cells, initiation of epithelial-to-mesenchymal transition, and changes in cell shape, movement, and attachment. Discovering potent drug cocktails capable of addressing the genetic and protein factors underlying pancreatic cancer’s development is formidable due to the disease’s intricate and varied nature. Method In this study, we introduce a fresh model utilizing Graph Attention Networks (GATs) to pinpoint potential drug pairings with synergistic effects for MNP, following the RAIN protocol. This protocol comprises three primary stages: Initially, employing Graph Neural Network (GNN) to suggest drug combinations for disease management by acquiring embedding vectors of drugs and proteins from a diverse knowledge graph encompassing various biomedical data types, such as drug-protein interactions, gene expression, and drug-target interactions. Subsequently, leveraging natural language processing to gather pertinent articles from clinical trials incorporating the previously recommended drugs. Finally, conducting network meta-analysis to assess the relative effectiveness of these drug combinations. Result We implemented our approach on a network dataset featuring drugs and genes as nodes, connected by edges representing their respective p-values. Our GAT model identified Gemcitabine, Pancrelipase Amylase, and Octreotide as the optimal drug combination for targeting the human genes/proteins associated with this cancer. Subsequent scrutiny of clinical trials and literature confirmed the validity of our findings. Additionally, network meta-analysis confirmed the efficacy of these medications concerning the pertinent genes. Conclusion By employing GAT within the RAIN protocol, our approach represents a novel and efficient method for recommending prominent drug combinations to target proteins/genes associated with pancreatic cancer. This technique has the potential to aid healthcare professionals and researchers in identifying optimal treatments for patients while also unveiling underlying disease mechanisms. Highlights Graph Attention Networks (GATs) used to recommend drug combinations for pancreatic cancer RAIN protocol applied to extract relevant information from clinical trials and literature Gemcitabine, Pancrelipase Amylase, and Octreotide identified as optimal drug combination Network meta-analysis confirmed the effectiveness of the drug combination on gene targets Novel and efficient method for drug discovery and disease mechanism elucidation Abstract Figure