Clustering of Drug Side Effects Based on Multi-Omics Data

The investigation of drug side effects is of paramount importance for identifying and predicting potential risks associated with pharmacological interventions, thereby enhancing patient safety and optimizing therapeutic outcomes. In-depth analysis of side effect data establishes a robust scientific foundation for drug optimization, personalized treatment strategies, and drug repurposing. With advancements in technology, researchers can now access multi-source biological data spanning diverse scales and levels. This study introduces the Integration of Multi-Omics Data for Drug Side Effect Clustering (IMOD-DSEC) model, an unsupervised learning framework designed to cluster drug side effects by integrating multi-omics data through a heterogeneous graph. The proposed model systematically generates meaningful node embeddings, harmonizes dimensional disparities, and facilitates effective clustering. Specifically, the model begins by generating semantically rich embeddings for each node based on its descriptive attributes, which are processed through a denoising autoencoder to ensure dimensional consistency. Two modules are then designed to aggregate multi-omics data through both intra- and inter-node type learning. The first module employs a Bi-LSTM model to refine the embeddings within nodes of the same type. The second module uses a random walk with restart strategy to sample highly correlated heterogeneous neighbor nodes for each central node, applying an attention mechanism to update the central node’s representation and produce the final embeddings. The k-means algorithm is subsequently used for clustering. Experimental results demonstrate the efficacy of IMOD-DSEC in accurately clustering multi-system drug side effects, underscoring its potential as a valuable tool for advancing drug safety research and multi-omics data integration.