Markov random field based method to predict side effects

Studying side effects of drugs is a very important issue in drug development. Based on the mechanism that drugs exert their side effects by interacting with molecules, we developed a Markov random field based method to predict drug side effects by evaluating the contributions of proteins to the side effects, which integrated information of the drug-side effect observation data, drug-target relations, and protein-protein interactions. Each protein was assigned a probability for its contribution to the side effect. With paralytic ileus as the example, 10 proteins with probabilities > 0.5 were identified. lozapine and amoxapine were successfully predicted to cause this side effect. The obtained probabilities of proteins were validated using the cross validation on drug-side effect relations with AUC 0.8. By implementing our algorithm on various side effects in SIDER database, proteins gaining high probabilities in many side effects were considered dangerous which may be unsuitable to be targets, which were validated in databases and literature. Developing a computational method to find these specific proteins responsible for a side effect will help us to understand the mechanism of this side effect, make side effect prediction, and develop safe and effective drugs.