We are constructing the "Percellome Database" containing many transcriptomes of mice exposed to a series of chemicals to elucidate the molecular mechanism of toxicity and to develop toxicity prediction technology. Acute toxicity of a chemical can be predicted to a certain extent by searching the similarity of the transcriptomes obtained by the single-dose exposure experiments. In addition, we are analyzing the relation between the transcriptome and the epigenome i.e. histone modification and genomic DNA methylation to understand the molecular mechanism of the repeated dose toxicity. We are attempting to expand the scale and improve the efficiency of the analysis by introducing artificial intelligence technologies. This approach should maximize the use of toxicogenomics technology for optimizing the experimental protocols for repeated dose toxicity studies towards 3Rs principle, and optimizing the process of in silico toxicity prediction by combining the available big data.