U.S. flag

An official website of the United States government

Dot gov

The .gov means it's official.
Federal government websites often end in .gov or .mil. Before sharing sensitive information, make sure you're on a federal government site.


The site is secure.
The https:// ensures that you are connecting to the official website and that any information you provide is encrypted and transmitted securely.


ICCVAM Biennial Report 2018-2019

Biennial Progress Report 2018-2019 Interagency Coordinating Committee on the Validation of Alternative Methods

Computational Tools Development

Computationally generated predictions of toxicity endpoints can inform decisions about testing priorities and sometimes eliminate the need for laboratory testing. ICCVAM agencies are developing tools to predict toxicity endpoints such as acute oral toxicity, inhalation toxicity, and endocrine disruption, as well as tools to relate in vitro data to in vivo toxicities through in vitro to in vivo extrapolation.
Collaborative Modeling Project for Androgen Receptor Activity (CoMPARA)

EPA established the worldwide CoMPARA consortium to develop computational approaches to screen chemicals for their potential androgenic activities. CoMPARA followed the approach used for CERAPP (Mansouri et al. 2016).

Regulatory requirements to screen chemicals for potential endocrine disrupting activity are being addressed via HTS approaches and computational modeling. CoMPARA (Mansouri et al. 2020) brought together scientists from 25 international groups, who contributed 91 predictive QSAR models to predict androgen receptor binding, agonist, and antagonist activity. Models were evaluated using literature data extracted from different sources and curated for quality. To overcome the limitations of single-model approaches, CoMPARA predictions were combined into consensus models that provided averaged predictive accuracy of approximately 80% for the evaluation set. These consensus models have been implemented into the free and open-source OPERA application to enable new chemicals of interest to be screened. The entire EPA DSSTox (Distributed Structure-Searchable Toxicity) database of ~750,000 chemicals was virtually screened. Predicted androgen receptor activities have been made available on the EPA CompTox Chemicals dashboard and in the Integrated Chemical Environment.

Collaborative Acute Toxicity Modeling Suite (CATMoS) Tool for Predicting Acute Oral Toxicity

CATMoS is a free online resource for screening organic chemicals for acute oral toxicity.

The ICCVAM Acute Toxicity Workgroup organized a global project to develop in silico models of acute oral systemic toxicity that predict five specific endpoints needed by regulatory agencies:

  • Very toxic (LD50 <50 mg/kg vs. all others)
  • Nontoxic (LD50 >2000 mg/kg vs. all others)
  • LD50 point estimates
  • Hazard categories under the EPA classification system (n=4) 
  • Hazard categories under GHS (n=5; Category 5 and Not Classified combined into a single category)

NICEATM invited scientists to develop models to predict any or all of these endpoints. NICEATM and the EPA National Center for Computational Toxicology (NCCT; now part of the EPA Center for Computational Toxicology and Exposure) collected a large body of rat acute oral toxicity data. Subsets of these data were used by project participants to build and test their models, and by NICEATM and the project organizing committee to evaluate the models. Project participants and potential users of the models attended a workshop at NIH in Bethesda, Maryland, in April 2018 (Kleinstreuer et al. 2018).

Models developed for the project that met both quantitative and qualitative criteria defined by the project organizing committee were used to generate consensus predictions for the acute oral toxicity endpoints of interest to regulatory agencies. The consensus predictions are available in CATMoS, which is implemented in v2.5 of OPERA, a free and open-source QSAR tool (Mansouri et al. 2018). The consensus predictions will also be available via the Integrated Chemical Environment and EPA's CompTox Dashboard. A journal article in preparation for submission in 2020 will describe generation of the consensus predictions.

Open Structure-activity/property Relationship App (OPERA)

OPERA is a free and open-source/open-data suite of QSAR models providing predictions for physicochemical properties, environmental fate parameters, and toxicity endpoints. OPERA is an ongoing collaboration between NICEATM and EPA.

QSAR models provide predictions of chemical activity that can augment non-animal approaches for predicting toxicity. However, the performance of QSAR models highly depends on the quality of the data and modeling methodologies used. The EPA National Center for Computational Toxicology (now part of the EPA Center for Computational Toxicology and Exposure) created OPERA to provide robust QSAR models for chemical properties of environmental interest that can be used for regulatory purposes (Mansouri et al. 2018). 

Additions to OPERA during 2018-2019 include predictions for:

  • Estrogenic activity from CERAPP (Mansouri et al. 2016).
  • Androgenic activity from CoMPARA (Mansouri et al. 2020).
  • Acute oral systemic toxicity from CATMoS (Kleinstreuer et al. 2018).
  • Physicochemical properties such as acid dissociation constant (pKa) and octanol-water dissociation coefficient (logD) (Mansouri et al. 2019).
  • Properties such as human plasma fraction unbound and hepatic intrinsic clearance that affect how substances behave in biological systems, to support an open-source workflow for IVIVE.
ICE Tools for IVIVE and Chemical Characterization

NICEATM created the Integrated Chemical Environment (ICE) to provide curated data and tools to facilitate the safety assessment of chemicals. Launched in 2017, ICE addresses the data needs frequently expressed by NICEATM stakeholders. The May 2019 ICE 2.0 update provided a new user interface to simplify searches and new tools that let users explore chemical properties and toxicity in more detail. Tools provided in the update included:

  • An update of the IVIVE tool (Casey et al. 2018) to provide more complex models, including those from EPA's high-throughput toxicokinetic (httk) package, that can improve prediction accuracy.
  • Chemical list characterization to explore physicochemical properties of chemical sets.
High-throughput In Vitro to In Vivo Extrapolation Using Tox21 Data

This NIEHS project developed and refined approaches to extrapolate all Tox21 chemical-concentration effect data to estimate human-equivalent exposure doses. The effort built on previous efforts using high-throughput toxicokinetics models and combined 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. NIEHS is continuing to refine the models for use in research and chemical screening prioritization. Additionally, a simple IVIVE workflow allowing users to select Tox21 assays and chemicals and extrapolate to estimated exposures is available in the NICEATM Integrated Chemical Environment (ICE). The ICE IVIVE tool provides three models with varied complexity and exposure routes: a one-compartment PK model, a three-compartment PK model, and a multi-compartment physiologically based PK model for oral and intravenous routes. These models use the ICE curated Tox21 data to predict equivalent administered in vivo doses for acute oral toxicity and endocrine disruption endpoints.

Optimization of Pharmacokinetic Models for In Vitro to In Vivo Extrapolation of Estrogenic Activity

NIEHS scientists evaluated and optimized IVIVE approaches using in vitro estrogen receptor activity to predict estrogenic effects measured in rodent uterotrophic studies. This work (Casey et al. 2018) evaluated the use of three PK models with varying complexities to extrapolate in vitro to in vivo dosimetry for 29 estrogen receptor agonists using data from validated in vitro and in vivo methods. In vitro activity values were adjusted using mass-balance equations to estimate intracellular exposure via an enrichment factor, and steady-state model calculations were adjusted using fraction of unbound chemical in the plasma to approximate bioavailability. Accuracy of each model-adjustment combination was assessed by comparing model predictions with lowest effect levels from guideline uterotrophic studies. The comparison found little difference in model predictive performance based on complexity or route-specific modifications. Simple adjustments, such as using the enrichment factor to account for in vitro intracellular exposure or fraction of unbound chemical to account for chemical bioavailability, resulted in significant improvements in the predictive performance of all models. The resulting computational IVIVE approaches accurately estimated chemical exposure levels that elicit positive responses in the rodent uterotrophic bioassay. Such studies are important for establishing confidence in the quantitative extrapolation of in vitro activity to relevant end points in animals or humans.

Semi-automated Extraction of Literature Data Using Machine Learning Methods

Identifying and extracting information from the full text of scientific publications is a critical step required in developing reference databases for establishing confidence in alternative approaches. However, manually extracting protocol details such as species, route of administration, and dosing regimen is labor-intensive and can introduce errors. NIEHS and the Department of Energy’s Oak Ridge National Laboratory are applying natural language processing and machine learning methods using both unsupervised and supervised approaches to identify specific data elements in the full text of scientific publications. For example, an unsupervised approach was developed to identify text segments (sentences) relevant to a set of criteria describing specific study parameters, such as species, route of administration, and dosing regimen. A binary classifier was then trained to identify publications that met the criteria. The classifier performed better when trained on the candidate sentences than when trained on sentences randomly picked from the text, supporting the hypothesis that this method could accurately identify study descriptors. This work is being expanded to include machine learning-based multivariate models combined with natural language processing to automatically extract text features that correspond to study descriptors and classify papers based on their adherence to minimum criteria derived from regulatory guideline studies. A publication is being drafted for submission in 2020.

Pred-Skin Web Portal for Predicting Human Skin Sensitizers

In collaboration with scientists at the University of North Carolina at Chapel Hill and the Universidade Federal de Goiás, NICEATM scientists contributed to the development of a model to predict skin sensitization potential of chemicals for two assays, human patch test and murine local lymph node assay, and implemented this model in a web portal (Braga et al. 2017). Work over the last two years focused on substantially revising and expanding the freely available web tool, Pred-Skin version 3.0, to integrate multiple QSAR models developed with in vitro, animal in vivo, and human ex vivo data into a consensus naïve Bayes model that predicts human effects. All models are freely accessible through the Pred-Skin v. 3.0 portal. A publication is being drafted for submission in 2020.


AOPXplorer is a tool to visualize non-animal assay data, mRNA levels, protein expression, and molecular biology data within an AOP network. The networks in AOPXplorer are built by U.S. Army scientists and the user community and are typically based on disease pathways and known modes of action for chemicals. During 2018 and 2019, several new pathways in AOPXplorer were developed and updated, including steatosis, signal transduction, regenerative proliferation, learning memory and cognitive decline, general cancer pathway, estrogen-mediated breast cancer pathway, PPAR-gamma lung fibrosis pathway, steroidogenesis, and skin sensitization. As of January 2020, AOPXplorer has been downloaded over 950 times. AOPXplorer is an app for the Cytoscape network visualization software and can be downloaded from the Cytoscape app store. Submissions of new pathways to AOPXplorer are welcomed.

Computational Neuronal Network Analysis for Extrapolation of iPSC Findings to Human In Vivo Effects

AFRL is developing an in silico mechanistic model of an in vitro experimental system consisting of human iPSC-derived dopaminergic neurons cultured on microelectrode arrays. While microelectrode array recordings of cultured neurons are useful for large, high-throughput experiments, cultured neurons form synaptic connections randomly rather than functionally as in the central nervous system, making it difficult to compare the firing activity recorded via microelectrode array to activity patterns produced by the human brain. This model should make it possible to extrapolate in vitro results to an expected impact on brain functions such as cognition and memory. An initial version of the model was implemented in the NEST modeling software. This combined platform contained neuronal types similar to those found in the human iPSC-derived neuronal cultures, parameterized using human and rodent data from the literature. Some key aspects of the behavior of the human iPSC-derived neurons could not be accurately represented due to limitations in the NEST software; the model was able to reproduce biological firing patterns qualitatively, but with unrealistically fast kinetics. The model is currently being redeveloped in the NEURON modeling software, which allows a more detailed implementation of neuronal mechanisms. This will allow the key behaviors of the human iPSC-derived neurons to be represented accurately.

Machine Learning Approaches to Assess New Threat Compounds

CCDC CBC is actively investigating machine learning approaches to assess new threat compounds. Specifically, the center is focused on three major initiatives focused on characterizing the toxicity of emerging threat compounds. These include:

  • Regression/classification models for assessing acute toxicity in non-human species.
  • Prediction of chemical targets using available in vitro data on human cell systems.
  • Prediction of chemical-target interaction with three-dimensional docking models.
Generalized Inhalation Toxicity Model

The AFRL 711th Human Performance Wing is collaborating with EPA to create an inhalation component for the R-based httk package. The goal of this effort is to develop a computational approach to generate high-throughput toxicokinetic estimates of chemical inhalation exposures. The inhalation component of httk will have three major modules: gas, aerosol, and mixture inhalation. In combination, these modules will address the most common occupational exposure scenarios. The project utilizes open-source software, which should encourage ongoing collaborations for updating and improving the models.

The gas inhalation module utilizes physicochemical property values from the OPERA QSAR tool to estimate the time course of blood and exhaled breath concentrations. A paper describing the gas inhalation module in more detail will be published in 2020 (Linakis et al. 2020). The aerosol module currently includes updated versions of two previously published lung deposition models to allow users more flexibility in determining which region of the lungs may be affected as a result of exposures to aerosols of a certain particle size. Upon completion and release of these modules, a mixture module will be pursued utilizing the current literature on gas-aerosol/particle interactions.

Expanded Decision Tree Software for Toxicity Classification

FDA is developing the Expanded Decision Tree software, which will be a free tool to classify compounds into six classes of relative toxic potential.

During the last seven decades, scientific advancements have led to an exponential increase in the number and types of chemicals to which humans are known to be exposed, leading to an ever-increasing need to screen and prioritize these substances according to their relative toxicity. The Cramer et al. (1978) Decision Tree is a screening and prioritization tool that sorts chemicals into three classes of relative toxicity. FDA updated and expanded the Cramer et al. Decision Tree to reflect the current state of the science and to make the decision tree applicable to a broader scope of substances, including those present in food, food contact materials, cosmetics, and dietary supplements. Additionally, FDA increased the number of classes of relative toxicity to six (non-toxic, low, medium, high, very high, and extreme toxicity) and quantified the toxic potential of each class by calculating a threshold of toxicological concern level for each class. By screening and prioritizing substances, the Expanded Decision Tree software will help focus resources on the safety assessments of substances with greater potential for public health risk and help reduce the use of animals for safety testing. An update on the development of the Expanded Decision Tree was presented at the September 2019 FDA workshop, “Implementing FDA’s Predictive Toxicology Roadmap: An Update of FDA Activities,” which highlighted activities to support and implement its Predictive Toxicology Roadmap.