Incorporating Prior Genomic Dose-Response Data to Support the Benchmark Dose Estimation of Toxicogenomics

Chemical risk assessment is an important tool to evaluate the toxicity of chemicals in the environment, and high throughput toxicogenomics plays an increasingly important role in risk assessment. In toxicogenomics, dose-response analysis for each gene is a data-limited situation, and thus parameter and benchmark dose (BMD) estimations typically have large uncertainty. To solve this problem, an informative prior by synthesizing toxicological information is integrated into the Bayesian benchmark dose modeling system (BBMD), a leading web-based toxicogenomics analysis application. We analyzed 276,126 toxicogenomics dose-response datasets and obtained plausible estimation of informative priors for seven commonly used continuous dose-response models. The effects of informative priors are investigated at the individual probe and pathway levels. Simulation studies based on six “true” models generated from typical genomic dose-response shapes show a significant decrease in uncertainty and an increase in accuracy of BMD estimates for most scenarios with informative priors than the counterpart with uninformative priors. The case study on the pathway analysis indicates that informative priors slightly improve the correlation between the pathway-based point of departure and apical point of departure. Overall, our study provides a practical strategy to incorporate existing toxicogenomic information as priors to improve the quality of chemical risk assessment. Graphic abstract