Prediction and principle discovery of drug combination based on multimodal friendship features

Abstract

Combination therapy, which can improve therapeutic efficacy and reduce side effects, plays an important role in the treatment of multiple complex diseases. Yet, the design principles of molecular combinations remain unclear. In addition, the huge search space of candidate drug combinations and the numerous heterogeneous data has brought us a big challenge. Here, we proposed a Friendship based Method (FSM), which integrates diverse drug-to-drug information to predict drug combinations for specific diseases. By quantifying the friendship-based relationship between drugs, we found that there is a moderate similarity between the drugs of effective drug combinations in a high-dimensional, heterogeneous feature space. Following this discovery, FSM applied a two-step strategy to predict clinically efficacious drug combinations for specific diseases. First, our method employs the friendship features to evaluate whether each drug is combinable. Then, the synergistic potential of combinable drugs was further evaluated. FSM was validated on two types of disease. The results show that FSM achieves substantial performance improvement over other state-of-the-art methods and tends to have low toxicity. These results indicate that our model could potentially offer a generic, powerful strategy to identify efficacious combination therapies in the vast search space.