A workflow for human health hazard evaluation using transcriptomic data and Key Characteristics-based gene sets

Key characteristics (KCs) are properties of chemicals that are associated with different types of human health hazards. KCs are used for systematic reviews in support of hazard identification. Transcriptomic data are a rich source of mechanistic data and are frequently interpreted through "enriched" pathways/gene sets. Such analyses may be challenging to interpret in regulatory science because of redundancy among pathways, complex data analyses, and unclear relevance to hazard identification. We hypothesized that by cross-mapping pathways/gene sets and KCs, the interpretability of transcriptomic data can be improved. We summarized 72 published KCs across 7 hazard traits into 34 umbrella KC terms. Gene sets from Reactome and Kyoto Encyclopedia of Genes and Genomes (KEGG) were mapped to these, resulting in "KC gene sets." These sets exhibit minimal overlap and vary in the number of genes. Comparisons of the same KC gene sets mapped from Reactome and KEGG revealed low similarity, indicating complementarity. Performance of these KC gene sets was tested using publicly available transcriptomic datasets of chemicals with known organ-specific toxicity: benzene and 2,3,7,8-tetrachlorodibenzo-p-dioxin tested in mouse liver and drugs sunitinib and amoxicillin tested in human-induced pluripotent stem cell-derived cardiomyocytes. We found that KC terms related to the mechanisms affected by tested compounds were highly enriched, while the negative control (amoxicillin) showed limited enrichment with marginal significance. This study's impact is in presenting a computational approach based on KCs for the analysis of toxicogenomic data and facilitating transparent interpretation of these data in the process of chemical hazard identification.