HUF-DCIE: A Hypergraph Based Unified Framework to Predict Drug Combination Integrated Effects
Liqiao Yan, Yi Cao · 2023
In recent clinical trials, taking multiple drugs simultaneously is a promising approach to treat the complex diseases. When using multiple drugs, it is necessary to consider the drug combination integrated effect, which consists of drug-drug interaction and drug combination synergy. Due to the enormous size of the combinatorial space, the drug combination integrated effect is hard to identify through wet-lab experiments. Deep learning methods have become an effective way to predict it recently, but most of them focus only on one aspect of the drug combination integrated effect, i.e., either drug-drug interaction or drug combination synergy, with few concentrating on both. We propose a unified framework, HUF-DCIE, to predict both aspects of drug combination integrated effect, with a hypergraph neural network that considers the heterogeneity of hypergraphs by learning hyperedge-dependent node embeddings. We evaluated HUF-DCIE on four representative datasets, including DrugBank DDI, Twosides, DrugComb, DrugCombDB. Our results show that HUF-DCIE outperforms four existing models in two scenarios, including random cross-validation and stratified cross-validation for drug combinations, thereby proving the validity of HUF-DCIE.