Machine learning approach to predict the adverse drug reaction in human

<p>In a new drug development process, a human safety is predicted from the results of non-clinical studies. However, unpredictable adverse events may have serious consequences such as withdrawal from clinical trials and the market. In particular, idiosyncratic adverse effects are rare and do not show a dose-dependence, so it is difficult to predict toxicity from non-clinical studies. In addition, it is clinically difficult to detect the idiosyncratic adverse effects because of the limited number of patients in clinical trial. Therefore, in the development stage of new drugs, prediction of candidate substances that may cause idiosyncratic adverse effects can be expected to improve the efficiency of clinical trials and post-marketing safety evaluations.</p><p>Recently, a large-scale medical information and artificial intelligence techniques are being available. Therefore, in silico methods for predicting the toxicity such as human adverse effects by utilizing chemical structural information and in vitro data have been developed. In silico prediction of idiosyncratic adverse effects can be expected to predict not only the toxicity of the candidate substance itself, but also the toxicity of various chemical substances such as impurities in the synthesis stage and human-specific metabolites.</p><p>In this symposium, we will introduce predictive model for severe skin adverse reactions such as SJS/TEN using JADER. Furthermore, we would like to introduce the development of predictive model for idiosyncratic drug-induced liver injury using chemical structural information and in vitro information.</p>