Modeling Polypharmacy Side Effects with Graph Convolutional Networks

Modeling Polypharmacy Side Effects with Graph Convolutional Networks

2018 | Marinka Zitnik, Monica Agrawal, Jure Leskovec
The paper "Modeling Polypharmacy Side Effects with Graph Convolutional Networks" by Marinka Zitnik, Monica Agrawal, and Jure Leskovec addresses the challenge of predicting side effects from polypharmacy, which is the concurrent use of multiple medications to treat complex diseases. Polypharmacy increases the risk of adverse side effects due to drug-drug interactions, which are often rare and not well understood. The authors introduce Decagon, a method that constructs a multimodal graph of protein-protein interactions, drug-protein target interactions, and polypharmacy side effects. Each side effect is represented as a different type of edge in the graph. Decagon uses a graph convolutional neural network to predict the exact side effect of a given drug combination. The model outperforms existing methods by up to 69%, accurately predicting side effects and learning representations that indicate co-occurrence of polypharmacy in patients. Decagon is particularly effective for side effects with strong molecular bases but also performs well for non-molecular side effects due to effective parameter sharing across edge types. The study opens new opportunities for using large pharmacogenomic and patient population data to prioritize polypharmacy side effects for further analysis.The paper "Modeling Polypharmacy Side Effects with Graph Convolutional Networks" by Marinka Zitnik, Monica Agrawal, and Jure Leskovec addresses the challenge of predicting side effects from polypharmacy, which is the concurrent use of multiple medications to treat complex diseases. Polypharmacy increases the risk of adverse side effects due to drug-drug interactions, which are often rare and not well understood. The authors introduce Decagon, a method that constructs a multimodal graph of protein-protein interactions, drug-protein target interactions, and polypharmacy side effects. Each side effect is represented as a different type of edge in the graph. Decagon uses a graph convolutional neural network to predict the exact side effect of a given drug combination. The model outperforms existing methods by up to 69%, accurately predicting side effects and learning representations that indicate co-occurrence of polypharmacy in patients. Decagon is particularly effective for side effects with strong molecular bases but also performs well for non-molecular side effects due to effective parameter sharing across edge types. The study opens new opportunities for using large pharmacogenomic and patient population data to prioritize polypharmacy side effects for further analysis.
Reach us at info@study.space
[slides] Modeling polypharmacy side effects with graph convolutional networks | StudySpace