Inferring protein domains associated with drug side effects based on drug-target interaction network

Background: Most phenotypic effects of drugs are involved in the interactions between drugs and their target proteins, however, our knowledge about the molecular mechanism of the drug-target interactions is very limited. One of challenging issues in recent pharmaceutical science is to identify the underlying molecular features which govern drug-target interactions. Results: In this paper, we make a systematic analysis of the correlation between drug side effects and protein domains, which we call pharmacogenomic features, based on the drug-target interaction network. We detect drug side effects and protein domains that appear jointly in known drug-target interactions, which is made possible by using classifiers with sparse models. It is shown that the inferred pharmacogenomic features can be used for predicting potential drug-target interactions. We also discuss advantages and limitations of the pharmacogenomic features, compared with the chemogenomic features that are the associations between drug chemical substructures and protein domains. Conclusion: The inferred side effect-domain association network is expected to be useful for estimating common drug side effects for different protein families and characteristic drug side effects for specific protein domains.