Application of data mining and visualization techniques for the prediction of drug-induced nausea in man

The therapeutic value of many drugs can be limited by gastrointestinal (GI) adverse effects such as nausea and vomiting. Nausea is a subjective human sensation, hence little is known about preclinical biomarkers that may accurately and effectively predict its presence in man. The aim of this analysis was to use informatics and data-mining tools to identify plausible preclinical GI effects that may be associated with nausea and that could be of potential use in its prediction. A total of 86 marketed drugs were used in this analysis, and the main outcome was a confirmation that nausogenic and non-nausogenic drugs can be clearly separated based on their preclinical GI observations. Specifically, combinations of common preclinical GI effects (vomiting, diarrhea, and salivary hypersecretion) proved to be strong predictors. The model was subsequently validated with a subset of 20 blinded proprietary small molecules and successfully predicted clinical outcome in 90% of cases. This investigation demonstrated the feasibility of data-mining approaches to facilitate discovery of novel, plausible associations that can be used to understand drug-induced adverse effects.