Our Percellome Toxicogenomics Project aims at describing dynamic and comprehensive toxicobiological gene network for the development of predictive toxicology using time- and dose-dependent transcriptomic responses induced by a chemical. A normalization method designated as “Percellome” was developed for microarrays and quantitative PCR to generate absolute copy numbers of each mRNA per one cell (in average) (Kanno J et al, BMC Genomics 7: 64, 2006). For this attempt, we adopted phenotype-independent approach, simply because not all the changes in mRNA expression can be anchored to the symptoms or overt phenotypes. We consider that this approach meet demand for food safety assessment as well. Unlike medicines, food ingredients rarely pre-specify the target molecule. Here, we briefly introduce our Project with an example on a flavor, estragole, which turned out to be a PPAR alpha agonist, like clofibrate. As a result of applying our toxicogenomics that enables elucidation of the molecular mechanism of toxicity and toxicity prediction to food ingredients, an unexpected target signal network has been clarified and toxicity prediction has become possible. Such a “phenotypic-independent” molecular toxicological and “comprehensive” approach may be useful in considering new pharmacology or the cooperation with pathology in the future.