Exploring drug combinations in a drug-cocktail network

Combination of different agents is widely used clinically to combat complex diseases with improved therapy and decreased side effects. It is necessary to understand the underlying mechanisms of drug combinations. In this work, we proposed a network-based approach to investigate drug combinations. Our results showed that the agents in an effective combination tend to have more similar therapeutic effects and more interaction partners in a drug-cocktail networks than random combination networks. Based on our results, we further developed a statistical model termed as Drug Combination Predictor (DCPred) by using the topological features of the drug-cocktail network, and assessed its prediction performance by making full use of a well-prepared dataset containing all known effective drug combinations extracted from the Drug Combination Database (DCDB). As a result, our model achieved the overall best AUC (Area Under the Curve) score of 0.92. Our findings provide useful insights into the underlying rules of effective drug combinations and offer important clues as to how to accelerate the discovery process of new combination drugs in the future.