Construction of a predictive model for evaluating multiple organ toxicity

The liver and kidneys are major target organs that suffer in adverse drug reactions, and liver and kidney toxicity are often present together. A multiple organ toxicological study is more helpful in understanding the effects of drugs in living systems than targeting a specific organ for a toxicity study. There are many prediction models for evaluating toxicity, but they are limited by single organ predictions and insufficient to understand the toxic mechanisms of drugs in the human body. Thus, we developed multiple organ toxicity prediction models and sought to lay a foundation for understanding the toxic effect of drugs on other organs, apart from the target organ. Here, we developed and evaluated the four computational prediction models (ANN, kNN, LDA, and SVM) that can predict whether a drug is liver toxic or liver-kidney toxic. To construct the predictive model, we extracted 210 molecular signatures of two classes of 108 drugs from TG-gate transcriptome data. Among the four algorithms, SVM was the ‘best’ method for multi-organ toxicity classification, with over 90% accuracy and the maximum power of classification with a small number of features. These bioinformatics tools will help researchers to recognize the side toxicity of drugs, not just in the target organ, before advancing them to clinical trials and exposing humans.