Carcinoscreen®: New short-term prediction method for hepatocarcinogenicity of chemicals based on hepatic transcript profiling in rats

Abstract

Carcinogenicity is one of the most serious toxic effects of chemicals, and highly accurate methods for predicting carcinogens are strongly desired for human health. Here, we developed a new prediction system named “CARCINOscreen®” for evaluating the carcinogenic potentials of chemicals using the gene expression profiles of liver tissues from rats after a 28-day repeated dose toxicity study. The prediction formula was generated using a support vector machine with predictive genes selected from 68 training chemical datasets; a predictive score was then calculated to predict the carcinogenic potentials of the tested chemicals. To ensure the accuracy of the prediction system, the chemicals were divided into three groups (Groups 1 to 3) according to the resulting hepatic gene expression profiles, and a prediction formula was generated for each group. The prediction system was capable of predicting the carcinogenic-ity of training carcinogens and non-carcinogens with an accuracy of 92.9% to 100%. The final prediction result was determined based on the maximum prediction value obtained with three independent prediction formulas to build up the CARCINOscreen®. The system was able to predict carcinogenicity accurately in 94.1% of the 68 training chemicals. An external validation trial was performed with 16 chemicals, consisting of various carcinogens targeting rat liver or other organs and non-carcinogens. The system identified 68.8% of all the chemicals and 100% of the rat liver carcinogens as carcinogens. Thus, the CARCI-NOscreen®, a novel system for predicting hepatocarcinogenicity, is a promising tool for the prediction of rat liver carcinogens.