Hierarchical Constrained Density Regression with Application to Genotoxicity Prediction

• Activity Number: 213 - Contributed Poster Presentations: Section on Risk Analysis
• Type: Contributed
• Date/Time: Tuesday, August 4, 2020 : 10:00 AM to 2:00 PM
• Sponsor: Section on Risk Analysis
• Author(s): Michael Pennell* and Matthew Wheeler
• Companies: Ohio State University and National Institute for Occupational Safety and Health
• Keywords: Count Data; Dirichlet-Process; Dose-Response Modeling; Nonparametric Bayes; Nonparametric Regression; Toxicological Risk Assessment

Abstract

Genotoxicity of chemicals is commonly tested using the Ames Assay. Although it is considered the gold standard, the Ames assay is unsuitable for high-throughput testing since it is time consuming and requires a large of amount of compound. Also, from a public health perspective, it would be advantageous to predict the results of an Ames assay prior to chemical development. Thus, to address these problems, we developed a Bayesian nonparametric model that can be used to predict features of a dose-response. Our model assumes that chemical-specific dose-response functions come from two different populations based on whether or not the function exhibits a particular feature (e.g., genotoxicity). Feature probabilities are modeled using a logistic regression dependent on chemical-specific predictors such as chemical properties or highthroughput assay results. The individual dose-response functions within each population are subject to an umbrella-ordering constraint and clustered using a Dirichlet Process, which provides further insight on relationships between chemicals. The model is used to predict the genotoxicity of 492 chemicals using genotoxic alert data.