Predicting unintended effects of drugs based on off-target tissue effects

Abstract

Unintended effects of drugs can be caused by various mechanisms. Conventional analysis of unintended effects has focused on the target proteins of drugs. However, an interaction with off-target tissues of a drug might be one of the unintended effect-related mechanisms. Therefore, we need to confirm this effect through quantitatively experimental analysis in this study. Because, there is no previous study about this effect so far. We have several purposes for this preliminary study. The first purpose is to confirm that how many tissues were overlap among all tissues in protein level in human body. The second purpose is to confirm drug response existence between target protein in tissue and no target protein in tissue. The third purpose is to confirm drug response difference in case of drug target proteins in other tissues. The fourth purpose is to confirm drug response difference in case of drug target proteins in same tissues. We propose two processes to predict a drug’s unintended effects by off-target tissue effects: 1) identification of a drug’s off-target tissue and

2) tissue protein - symptom relation identification (tissue protein - symptom matrix). The first part is to identify the off-target tissue of the drug of interest. We utilized the ATC code and target proteins of the drugs to identify the off-target tissue. The second part is to find relations in the tissue protein - symptom matrix (T-SM). Our T-SM has relations between tissue proteins and symptoms. We can predict the unintended effects of a drug of interest by the off-target tissue effect from confirming the tissue protein on T-SM. The tissue protein-symptom matrix had 5,338 relations. Those tissue proteins were combined from 78 tissues and 204 drug target proteins. The target protein proportion in our proposed matrix was 15% of all known drug target proteins. Total 58 symptoms including headache, apnea, diarrhea, seizures, and so on were used as the other dimension on T-SM. Drug information had key roles in making the relations of the matrix. Drug information contributed to three of the total five relations to create a relation of T-SM. The drug information was collected from 247 marketed drugs. Using this method, we predicted that 1,177 (10.7%) side-effects were related to off-target tissue effects in 11,041 known side-effects. Off-target tissues and unintended effects of successful repositioning drugs were also predicted. We evaluated the tissue protein - symptom relations of T-SM by confirming the relations through literature mining. A tissue protein - symptom relation on T-SM contains two pieces of linkage information. One of the linkages is tissue - symptom relation, and the other is protein - symptom relation. We confirmed a conventional co-occurrence of two entities from each relation in one sentence in the abstracts of published papers. We predicted unintended effects of drugs as well as those effect-related off-target tissues. By using our prediction, we are able to reduce drug side-effects on off-target tissues and provide a chance to identify new indications of drugs of interest.