Biostatistics & Computational Biology Branch
BCBB investigators involved in Toxicology/Carcinogenesis projects, often in collaboration with scientists in the National Toxicology Program (NTP) , include: Benedict N. Anchang, Pierre R. Bushel, John House, Nicole Kleinstreuer, Alison A. Motsinger-Reif, Keith R. Shockley, David M. Umbach, Matt Wheeler.
Rodent Toxicology and Carcinogenesis: The NTP uses rodent studies to predict the toxicity and/or carcinogenicity of chemicals to humans. A typical study design evaluates animals in a zero-dose control group and in three test groups at increasing doses of a test substance. Routine endpoints involve clinical pathology and tissue histopathology. Additional assessments may include sperm motility, genotoxicity, and toxicogenomics. The BCBB has previously established data-analytic approaches that detect toxicologically relevant changes in these endpoints while controlling false-positive error rates and minimizing false negatives. We are currently developing methods to detect trends in binary tumor response that account for treatment-induced lethality and litter structure. To facilitate wide use of these methods, we are creating a comprehensive toxicology analysis suite (ToxicR) that will be available as an R package as well as a web-app. This software will allow researchers easy access to traditional toxicologic analyses including dose-response modeling, visualization, and benchmark dose analyses.
Chemical Mixtures: Human exposure to environmental chemicals involves multiple compounds together; consequently, the study of co-exposures or mixtures encompass one of the goals of the NIEHS Strategic Plan. The NTP studies both defined mixtures, those whose component chemicals are fully known, and complex mixtures, those whose components are incompletely known (e.g., botanical supplements). With defined mixtures, a common task is evaluating whether the dose-response relationship for a mixture is predictable from the individual dose-response relationships of components under various additivity assumptions. We are refining statistical methods that evaluate additivity to enhance their robustness to misspecification of the dose-response models used for component chemicals. Safety evaluation of complex mixtures, like botanicals, is difficult due to inconsistency in composition among preparations from different sources. Consequently, regulators face the problem of determining which preparations are “sufficiently similar” based on various biological and chemical endpoints. “Sufficient similarity,” though having commonsense appeal, lacks an operational definition. We are addressing this challenge by developing novel techniques that combine dimension reduction and clustering to delineate groups of mixtures that share toxicological characteristics.
High Throughput Screening (HTS): Thousands of chemicals currently in wide commercial use have never been tested for adverse effects on humans or the environment. Quantitative high throughput screening (qHTS) assays are multiple-concentration experiments that can simultaneously evaluate the activity of thousands of chemicals. These experiments can be used to identify and prioritize candidates for toxicity testing and can reveal potentially harmful compounds through predictive modeling. We have contributed both to the experimental design of qHTS experiments and to large-scale data mining efforts across multiple assays. In addition, we perform genome-wide association studies in high-throughput screening of anticancer drugs to better understand the genetics of drug response and toxicity. The BCBB has also developed tools, applicable to qHTS data, to address various goals such as: quality control, chemical-assay interference, identification of active compounds, estimation of compound potency, ranking compounds by overall activity level, and prediction of in vivo toxicity from in vitro data (e.g. by target organ). We also developed a method for large-scale nonlinear dose-response modeling that employs an evolutionary algorithm which simultaneously optimizes functional form and parameter estimates.
In addition to analyzing the data one chemical at a time, more sophisticated quantitative structural activity relationship (QSAR) models can incorporate hundreds of chemicals with their associated dose-response curves. We developed methodologies that learn the chemical activation dose-response surface of HTS assays based upon QSAR information. Modeling the activation surface allows prediction of the dose-response relationship for chemicals that were not tested. Current work extends this approach to multiple HTS assays, which accounts for the dose-response correlation between assays as well as QSAR information.
Toxicogenomics: Toxicology, like most of the biomedical sciences, has benefited from the rapid development of genomic technologies. We contribute to the design and conduct of whole genome and targeted gene expression analyses. As part of these efforts, we develop and apply computational tools to study differential gene expression and transcriptomic dose-response. Our approaches can accommodate experiments with primary cells, immortalized cells, induced pluripotent stem cells, and tissue-on-a-chip. These expensive high-throughput and multi-factorial experiments may include more than 1000 treatment/sample combinations in a single analysis. We have developed pipelines for transcriptomic data analysis and visualization of such designs and made them accessible through user-friendly tools such as R Shiny applications. In addition, we have designed and implemented bioinformatic workflows to detect single nucleotide variants (SNVs) from whole exome sequencing of DNA; and we have applied them to studies of mice with hepatocellular carcinomas arising spontaneously or due to chronic exposure to chemicals.
Consideration of the dynamic effects of cellular heterogeneity during normal and disease development presents a novel opportunity to advance both predictive toxicology and precision medicine. The new ability to measure high resolution molecular phenotypes directly from individual cells obtained from patient or animal samples allows researchers to objectively study the mechanisms underlying tumor progression and toxicity. Accordingly, we are developing dynamic and spatial network-based computational models that integrate single-cell data and resolve key patterns in the data. We plan to expand and verify our results through strategic implementation of experimentation and bioinformatic analysis.
The branch is also actively involved in assisting the development of. BMDExpress is an application that analyzes dose-response data in differential gene expression experiments, and it is considered the gold standard for estimating chemical toxicity and points of departure from such experiments. Our work includes increasing computational efficiency and stability of the analysis, as well as developing novel model averaging and Bayesian non-parametric pipelines to analyze this data more efficiently.