Data analysis pipelines are being developed for Tox21 Phase II quantitative high-throughput screening data to determine the activity of compounds in assays.
The developed ranking, or calling procedure, accounts for compound potency, efficacy, and data reproducibility. A manuscript describing this pipeline was published, and computational tools to analyze Tox21 data have been made publicly available through the Tox21 Toolbox
on the NTP public website.
A prioritization approach is being developed that includes compounds showing clear evidence of activity in the quantitative high-throughput screening genotoxicity assays and compounds that are weakly active based on chemical structure-activity relationship analysis.
A manuscript using this approach as part of the analysis of the Tox21 quantitative high-throughput screening p53 activation assay was published in FY 2017. In FY 2018, a more extensive modeling exercise was completed using data from all Tox21 quantitative high-throughput screening assays.
This project entails the development of two graphical user interfaces for viewing Tox21 data. One graphical user interface is used to explore the concentration-response data in a line chart, and the second is used to explore compound similarity relationships in terms of their chemical structures and activities in Tox21 quantitative high-throughput screening assays.
Prototype graphical user interfaces were first developed during FY 2013 and made public in FY 2015. In FY 2018, these computational tools to analyze Tox21 data were developed further and expanded and made publicly available through the Tox21 Toolbox
on the NTP public website.
This unsupervised data analysis is focusing on methods (data organization based on patterns and performed by software) to identify chemicals that exhibit biological properties similar to those of well-characterized toxicants from the quantitative high-throughput screening assays used to screen the 10K library.
The results are being used to help prioritize compounds for more extensive toxicological testing. An updated web interface for multiple integrated tools was made public in FY 2018, along with a supporting manuscript, and further updates were begun for release in FY 2019.
Basic machine learning workflows using the Tox21 data for model building and data exploration are available on ICE
This project entails the development and refinement of ways to extrapolate all Tox21 chemical-concentration effect data to estimated human equivalent exposure doses. The effort builds on previous efforts (outside of NTP) using high-throughput toxicokinetics models and combines them with in silico-estimated parameters.
A publicly available web application based on these methods is available through the Tox21 Toolbox
on the NTP public website, and refinement and use of the models in a research and prioritization context continued throughout FY 2018. Additionally, a simple IVIVE workflow allowing users to select Tox21 assays and chemicals and extrapolate to estimated exposures is available on ICE
This established Tox21 cross-partner project is developing and using a data-driven approach to choose cells to maximize biological diversity. A content maximization approach, along with programmatic interest, were used to pick a diverse set of 30+ cells based on publicly available gene expression data.
Baseline gene expression profiling is planned in these cells using the appropriate technology platform, which will allow harmonization across projects. In vitro chemical testing will commence subsequently.
This project involves the comparison of Tox21 data analysis methods, identification of higher-confidence chemical-assay actives, and development of a website for public access to the data and visualizations. A web interface with the aggregated hit-call function was developed during FY 2018 for an anticipated in FY 2019.
Under this project, bioinformatics pipelines are being developed for genomic and transcriptomic gene expression and mutational analysis on a genome-wide level using next-generation sequencing technologies to build signatures of toxicity and chemical exposure.
The effort was recently expanded to evaluate gene expression changes in frozen tissue samples from brain subregions obtained from genetic toxicity studies conducted as part of the NTP Cell Phone Radio Frequency Radiation study. In addition, new informatic tools have been developed to better identify long noncoding RNAs.
This project is supporting future validation of high-throughput in vitro test methods and in silico models of estrogenic activity. NICEATM created a comprehensive database of high-quality in vivo data from over 1,000 scientific articles describing uterotrophic assay experiments for more than 2,660 distinct combinations of chemicals, studies, and protocols.
These data have potential utility for developing adverse outcome pathways or models of estrogenic activity, prioritizing chemicals for further testing, or evaluating species-specific responses to chemicals.
The database is described in a manuscript
published in FY 2016. This activity is complete, but the database is being used as a training set for the automation of systematic reviews.
NICEATM developed and applied one-compartment or physiologically based pharmacokinetic models to data from validated in vitro (EPA ToxCast estrogen receptor pathway model) and in vivo (uterotrophic) methods to correlate in vitro and in vivo dosimetry quantitatively for estrogen receptor reference chemicals.
This approach highlighted the importance of pharmacokinetic considerations in assessing and ranking endocrine-active chemicals based on in vitro, high-throughput screening assays.
Subsequent work has focused on understanding the effect of various parameters, such as using free plasma concentration as a surrogate for total plasma concentration and comparing multiple modeling approaches and was published in FY 2018.
This project is determining the potential of high-throughput screening data to reduce animal use for acute oral toxicity testing. NICEATM analyzed high-throughput screening data from Tox21 and ToxCast for correlation and model fitting to rat oral LD50 data to determine which tests or combinations of tests best characterize the rat oral toxicity data.
The analysis suggests that combinations of in vitro assays and data from small model organisms, such as zebrafish, offer promise for predicting outcomes of rat acute oral toxicity tests.
A global collaboration to use quantitative structure-activity relationships, Tox21, and other alternative sources to predict acute oral toxicity has been initiated. Consensus models have been built as a result of the collaboration and a manuscript is being drafted.
This evaluation of various in silico methods for predicting the extent of xenobiotic metabolism is also identifying metabolites and for prioritizing chemicals in the Tox21 10K library.
Compuional methods are used to partition the 10K library and develop subsets of chemicals that are likely to be metabolized appreciably in humans.
This project includes identifying patterns of exposure-induced biological responses to characterize toxicity and disease pathways and facilitate extrapolation of findings from model species to humans. Criteria were developed for selecting the best target set of genes representing humans, rats, mice, and zebrafish.
A manuscript describing this work was published in FY 2018. A manuscript describing the gene selection process for the creation of a zebrafish S1500+ gene set is in preparation for submission in early 2019.
This integration of data from nine Tox21 and ToxCast assays into a computational model will be used to predict agonist and antagonist activity against the androgen receptor pathway.
A manuscript describing this work was published in FY 2017. A follow-up manuscript comparing the results of the computational model against reference chemicals derived from an in vivo database was published in FY 2018.
Using the computational model of the androgen receptor pathway, NICEATM developed QSAR models to predict androgen receptor binding and activity.
These QSAR models are currently being refined, with a goal of using them to predict androgen receptor pathway activity of chemicals in the EPA Endocrine Disruptor Screening Program. A manuscript is expected to be submitted in late 2018 or early 2019.
To develop a reference chemical list for in vitro androgen receptor binding and transactivation assay activity, NICEATM conducted literature reviews to identify information about in vitro androgen receptor binding and transactivation assays for 127 putative androgen-active or androgen-inactive chemicals.
The final database
was made available to the public on the NTP website and manuscript describing this work was published, both in FY 2017. A parallel data curation effort with OECD and EPA partners focused on in vivo androgen activity data.
These data will be used for evaluating high-throughput screening approaches, testing strategies, and further development of alternative test methods. A manuscript describing this work was published in FY 2018.
NICEATM is mapping Tox21/ToxCast assay targets to known modes of action for developmental toxicity, acute toxicity, and carcinogenicity. Carcinogenicity assay target mapping is informed by the key characteristics of carcinogens and the hallmarks of cancer.
Mapped assays are being combined with in silico features to build Bayesian network models to provide probabilistic predictions of chemical hazard. A manuscript describing this approach is being drafted.
NICEATM has curated high-throughput screening data from Tox21 and EPA ToxCast HTS program to identify and exclude low-confidence activity calls. Factors considered in the curation include chemical stability and purity information, robustness of concentration-response curve fits, and contextualization of active concentrations relative to testing range.
These data are available through the NICEATM ICE database. ICE, launched in 2017, provides high-quality, curated data from NICEATM and its partners as well as other data resources and tools to support development of new approaches for assessing chemical safety.
To better characterize chemicals identified in Tox21 quantitative HTS assays as having farnesoid X receptor alpha agonist or antagonist activity, NICEATM and collaborators evaluated them using four experimental approaches.
Experiments generally confirmed the Tox21 results, provided orthogonal data on protein-to-protein interactions and receptor docking, and translated those results to an in vivo system (larval medaka assay).
The study, presented at the 2018 SOT meeting, demonstrated an approach to targeted evaluation of putative bioactivity derived from HTS data.
Methods and software were developed for performing genomic dose-response analysis to identify sensitive, screening-level potency estimates. An expert panel meeting to discuss the proposed method was held and software released in FY 2017.
A manuscript describing NTP’s method and approach, along with a manuscript describing the software, were published in FY 2018.
This evaluation focused on determining if dose-response modeling of toxicogenomics data from short-term, in vivo studies can be used to identify biological effect points of departure that are comparable in potency to those derived from long-term toxicity studies.
A manuscript describing the findings from these studies will be published in FY 2019.
The goal of this project was to develop semi-automated approaches to identify reference chemicals for Tox21/ToxCast assay targets with EPA.
A manuscript on “RefChemDB” was submitted in FY 2018. A manuscript is being drafted with NCATS and EPA co-authors on the analysis of Tox21 assays designed to measure luciferase inhibition and autofluorescence to identify interference reference chemicals and build structure-based models to predict assay interference; publication is expected in FY 2019.