Training set

Training sets (download here) for five adverse drug effects (ADEs): myocardial infarction, arrhythmia, cardiac failure, hepatotoxicity and nephrotoxicity, were created based on information from drug labels. ADEs can be described in different sections of the label: “Boxed warning” (BW), “Warnings and Precautions” (WP) and “Adverse reactions” (AR). The first two sections usually contain the most severe and frequent ADEs which have causal relationships to the drug; therefore if ADE is described in BW or WP section, or drug was withdrawn from the market because ADE, then the drug is considered as “active” in the corresponding training set. All other drugs in training sets are considered as “inactives”. In order to filter out false negatives we included into the training sets only “inactive” compounds with systemic routes of administration such as oral, intravenous and intramuscular ones. Drugs with unclear relationships to ADE were excluded from the datasets.

Myocardial infarction. The training set contains 896 compounds with 92 active drugs causing myocardial infarction, mentioned in BW or WP sections of labels or withdrawn from the market because myocardial infarction.

Cardiac failure. The training set contains 904 compounds with 83 active drugs causing or exacerbating cardiac failure, mentioned in BW or WP sections of labels.

Arrhythmia. The training set contains 911 compounds with 177 active drugs causing qt interval prolongation or ventricular tachycardia, mentioned in BW or WP sections of labels, withdrawn from the market because arrhythmia or mentioned in CredibleMeds website (https://crediblemeds.org/), which contains lists of drugs associated with arrhythmia.

Nephrotoxicity. The training set contains 900 compounds with 98 active drugs causing acute kidney failure resulted from severe nephrotoxicity, mentioned in BW or WP sections of labels or withdrawn from the market because nephrotoxicity.

Hepatotoxicity. The training set contains 684 compounds with 181 active drugs causing acute liver failure resulted from severe hepatotoxicity, mentioned in BW or WP sections of labels, withdrawn from the market because hepatotoxicity, or mentioned in LiverTox website (https://livertox.nih.gov/) as causing severe liver damage with acute hepatic failure. This dataset contains “inactive” drugs with only oral mode of administration and average daily doses more than 10 mg, because intravenous and intramuscular drugs as well as oral drugs with lower daily doses are rarely cause severe liver damage because insufficient concentration in the liver.

Structure-activity relationships were built for five training sets using PASS (Prediction of Activity Spectra for Substances) software (http://www.way2drug.com/PASSonline).