You are here:
Modeling Pathway Interactions using High-Throughput Transcriptomics Data
Citation:
Judson, R., I. Shah, Woodrow Setzer, D. Haggard, T. Sheffield, AND J. Harrill. Modeling Pathway Interactions using High-Throughput Transcriptomics Data. Presented at Society of Toxicology (SOT) annual meeting, Baltimore, MD, March 10 - 14, 2019.
Impact/Purpose:
New experimental methods for running whole genome high-throughput transcriptomics (HTTr) experiments are making it possible to simultaneously assess many chemicals in dose response to aid both hazard identification and potency estimation. Here we describe results of a pathway modeling effort in which over 2000 chemicals were run in an HTTr screen in MCF-7 cells using the BioSpyder TempO-Seq platform (8 test concentrations / chemical).
Description:
High-throughput in vitro methods are being increasingly used for assessing the safety of chemicals. Two aspects of hazard assessment are hazard identification (i.e., what targets, pathways or processes does a chemical perturb) and potency estimation (i.e., at what concentration or dose does the chemical perturb these biological processes?). New experimental methods for running whole genome high-throughput transcriptomics (HTTr) experiments are making it possible to simultaneously assess many chemicals in dose response to aid both hazard identification and potency estimation. Here we describe results of a pathway modeling effort in which over 2000 chemicals were run in an HTTr screen in MCF-7 cells using the BioSpyder TempO-Seq platform (8 test concentrations / chemical). The raw count data was processed using the DESeq2 R-package to generate log2 fold change data for each chemical-concentration sample, for each of 3 biological replicates. Multiple pathway-level concentration response modeling methods were compared. These include variants of GSEA (Gene Set Enrichment Analysis), and a simple comparison of the average fold changes of genes in and out of a pathway (In-Out). Different filters for significance were used (filtering at the gene vs. pathway level). Concentration-response modeling was performed using the ToxCast Pipeline (tcpl) software. Method performance was evaluated using reference chemicals for multiple molecular targets, curated from the literature into a database called RefChemDB. Three main results are (1) The simple In-Out and the GSVA (Gene Set Variation Analysis) variant of GSEA were able to detect appropriate reference chemical / pathway interactions with potencies similar to what is seen for in vitro target-based assays; (2) For many chemicals, many pathways are activated in a non-specific way, indicating overall cell stress; and (3) Further research will be required to select an optimal method for assessing pathway level potencies from HTTr data. This abstract does not necessarily represent U.S. EPA policy.