In their March 2026 Society of Toxicology paper, “A workflow for human health hazard evaluation using transcriptomic data and Key Characteristics-based gene sets”, Tsai et al. propose a framework to evaluate transcriptomic data through two paradigms: Key Characteristics (KCs) of chemical compounds - expert-defined properties of chemicals associated with specific human health hazards - and pathway annotation databases. The authors first consolidated 72 individual KCs from seven published hazard-specific sets (covering carcinogens, cardiovascular toxicants, endocrine disruptors, reproductive toxicants, hepatotoxicants, and immunotoxicants) into 34 non-redundant umbrella KC terms. They then systematically mapped Reactome and KEGG pathways to these terms and generated parallel “KC gene sets" derived from Reactome and KEGG for each umbrella KC term by pooling all genes contained in the mapped pathways.

The Reactome- and KEGG-derived KC gene sets showed low gene overlap (most Jaccard scores below 0.1), confirming that the two databases are largely complementary rather than redundant and suggesting that optimal results can be obtained by using both gene sets in parallel. Reactome KC gene sets covered 77% of Reactome's annotated human genes.

The proposed workflow was then validated across four compounds. Enrichment analysis correctly identified immunotoxicity and oxidative stress for benzene, hepatotoxicity and carcinogenicity for TCDD (including strain-specific differences in AHR-driven liver fibrosis), and cardiac and mitochondrial dysfunction for the known cardiotoxicant sunitinib, while the non-cardiotoxic antibiotic amoxicillin showed minimal enrichment, as expected. GSEA and ORA were complementary, with GSEA showing greater sensitivity overall. Overall, the results suggest the proposed workflow may be a useful tool for systematic integration of transcriptomics into chemical hazard assessment.