Adverse drug effects (ADEs) are one of the leading causes of death in developed countries, and the main reason for drug recalls from the market. The ADEs associated with action on the cardiovascular system are the most dangerous and widespread. Treatment of human diseases often requires the intake of several drugs, which can lead to drug-drug interactions (DDIs) causing an increase in the frequency and severity of adverse effects. Evaluation of ADEs, as well as the effect of DDIs on their manifestation, is a non-trivial task and requires numerous experimental and clinical studies. To solve this problem, we developed a computational approach to assess the cardiovascular effects of DDIs. This approach is based on the analysis of FDA spontaneous reports (SRs) to identify DDIs with subsequent creation of structure-activity relationships (SARs) for pairs of drugs to predict five cardiovascular ADEs: myocardial infarction, ischemic stroke, ventricular tachycardia, cardiac failure, and arterial hypertension. At the first stage of our approach, we applied L1-regularized logistic regression to SRs for the identification of pairs of drugs that interact with each other and cause ADEs more frequently than individual drugs (examples of synergy and additivity, “actives” in SAR models). Using the same method we also identified pairs of drugs that do not interact with each other (“inactives” in SAR models). At the second step, five SAR models were created based on the obtained information. We used probability estimates for 1554 human target calculated by PASS Targets software for each drug to create descriptors. Sum and difference of estimates were used as descriptors for drug pairs. To create SAR models we also used L1-regularized logistic regression. Accuracy values were calculated based on 5-fold cross-validation procedure. The obtained datasets include on average 3500 drug pairs with active/inactive ratio 1:3. The average area under the ROC curve of obtained SAR models was 0.9, and the average balanced accuracy was 0.82. The predicted drug targets, which were taken as descriptors, can also be used to hypothesize the mechanisms of cardiovascular ADEs of DDIs. The created five SAR models can find practical application in the clinic for the selection of the safest combinations of drugs.

In silico assessment of cardiovascular adverse effects of drug-drug interactions

2019 Spring National Meeting

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.