Systematic linkage of transcriptomics data to Adverse Outcome Pathways for toxicity prediction

Authors: Sidra Adil^A, Theo de Kok^A, Egon Willighagen^A, Tooba Abbassi-Daloii^A, Marvin Martens^A

^ADepartment of Translational Genomics (TGX), NUTRIM, Maastricht University, Maastricht, Netherlands

Adverse Outcome Pathways (AOPs) are structured frameworks that describe how chemical exposure can lead to adverse health effects via so-called Key Events (KEs). To improve their use in risk assessment, we explored how transcriptomics data can be systematically linked to AOPs. Our goal was to enrich AOPs with genesets to better connect molecular changes to toxicological outcomes. Based on the AOP-Wiki, we generated extensive AOP networks specific to liver, brain, kidney, and lung toxicity. For KEs lacking ontological annotations, we systematically mapped them to biological process terms from Gene Ontology (GO) using AMiGO2, and then linked these to corresponding genesets. We tested the approach by using a publicly available Open TG-GATES human liver dataset exposed to ethanol at two concentrations and two time points. This dataset was used to quantify KE activity within the liver AOP network. This was done by following the single-sample gene set enrichment analysis (ssGSEA) approach. The AOP networks for liver, brain, kidney, and lung comprise 195, 222, 68, and 142 KEs, respectively. In total, 56.7% of KEs were already annotated with biological process terms, and we added annotations for the remaining unannotated ones. In the liver network, genesets were added for 149 out of 195 KEs (76%). Using these KE genesets, we calculated KE activity scores as a result of ethanol exposure. The interpretation of these results is ongoing. Our research supports predictive toxicology by predicting early molecular changes that precede adverse outcomes. Despite remaining challenges such as incomplete annotation and manual curation requirements, this work demonstrates a systematic strategy for integrating transcriptomics with AOP frameworks. By linking KEs to functional genesets and calculating their activity in response to chemical exposure, this research allows for more mechanistic, data-driven chemical risk assessments.