Presentation for Texas A&M Superfund Research Center virtual learning series, Big Data in Environmental Science and Toxicology. More details at https://superfund.tamu.edu/big-data-session-2-aug-18-2021/
4. Center for Computational Toxicology and Exposure, US-EPA, RTP, NC
http://www.orcid.org/0000-0002-2668-4821
New Approach Methods
What is That???
The views expressed in this presentation are those of the authors and do not necessarily reflect the views or policies of the U.S. EPA
Antony John Williams
williams.antony@epa.gov
Thanks to John Cowden, Katie Paul-Friedman, Grace Patlewicz and John Wambaugh
6. • National Center for Computational Toxicology
established in 2005 to integrate:
– High-throughput and high-content technologies
– Modern molecular biology
– Data mining and statistical modeling
– Computational biology and chemistry
• Staffed by ~60 employees as part of EPA’s Office of
Research and Development
• Home of ToxCast & ExpoCast research efforts
• Key partner in U.S. Tox21 federal consortium
The big jump was the National Center for
Computational Toxicology
8. What’s a NAM?
• NAM = New Approach Methodologies
• Commonly defined to include in silico
approaches, in vitro assays, as well as the
inclusion of information from the exposure of
chemicals in the context of hazard and
exposure assessment.
• Defined in the EPA’s TSCA Alternative Toxicity
Strategy as:
• a broadly descriptive reference to any technology,
methodology, approach, or combination thereof that
can be used to provide information on chemical
hazard and risk assessment that avoids the use of
intact animals.
https://echa.europa.eu/documents/10162/22816069/scientific_ws_proceedings_en.pdf
https://www.epa.gov/assessing-and-managing-chemicals-under-tsca/alternative-test-methods-and-
strategies-reduce
Thanks: John Cowden
9. Examples of New Approach Methodologies
• In silico (e.g. QSAR and Read-across)
• Estimate effects and doses
• Consensus exposure modeling
• In vitro assays
• Broad / screening (transcriptomics, cell painting)
• Targeted (receptors, enzymes)
• In vitro PODs, modes / mechanisms of action
10. Examples of New Approach Methodologies
• In vitro Toxicokinetics
• Allow conversion of an in vitro POD to in vivo (IVIVE)
• High-throughput Exposure Measurements
• To fill data gaps in monitoring data
• Computer models
• Hazard models to integrate multiple in silico and in vitro data streams
• Exposure models to increase information on different pathways of exposure
11. The Dashboard and NAMS
• Our overview of NAMS will specifically relate to the availability
of data in the CompTox Chemicals Dashboard you can access
• All data is accessible at http://comptox.epa.gov/dashboard
• See the FIRST session for details
16. Different QS(X)R predictions
• There are many different “QSAR-related” predictions available
• QSPR: quantitative structure–property relationships
• QSAR: quantitative structure–activity relationships
• QSUR: quantitative structure-use relationships
17. Experimental and Predicted Data
From the LAST SESSION
• Physchem and Fate & Transport
experimental and predicted data
• Data can be downloaded as Excel, TSV
and CSV files
• Predictions: multiple algorithms
• EPI Suite: Estimation Program Interface
• ACD/Labs (commercial)
• TEST: Toxicity Estimation Software Tool
• OPERA: OPEn structure–activity/
property Relationship App
19. TEST and OPERA Predictions
DESKTOP tools if you need them
20. Property and Fate and Transport Data
~25 MILLION pre-predicted values
• We have built QSPR models based on tens of thousands of
property data points curated over the past decade
• We push our “QSAR-Ready” chemical structures through
predictions to produce property predictions
29. What do we use predictions for?
• Property predictions can be used for
• Experimental design – what chemicals are too volatile to
perform bioactivity screens on?
• As inputs to other models – for example toxicokinetic models,
exposure models (we cannot measure all the properties we
need to build models)
• To use as flags related to persistence and bioaccumulation
30. Machine Learning NAMS
Chemical Structure
and Property Descriptors
humectant
lubricating
agent
perfumer
pH
stabilizer
oxidizer
heat
stabilizer
photo-
initiator
masking
agent
hair dye
organic
pigment
flavorant
flame
retardant
film
forming
agent
foam
boosting
agent
foamer
reducer
rheology
modifier
skin
protectant
skin condi-
tioner
soluble
dye
catalyst chelator colorant crosslinker emollient emulsifier
fragrance
plasticizer
monomer
solvent
antistatic
agent
anti-
oxidant
anti-
microbial
adhesion
promoter
additive
for rubber
additive
for liquid
system
whitener
wetting
agent
viscosity
controlling
agent
vinyl
UV
absorber
ubiquitous
surfactant
pre-
servative
oral care
hair condi-
tioner
emulsion
stabilizer
buffer
additive
Probabilistic
Predictions of
Potential Chemical
Uses
Chemical Functional Use Database (FUSE)
Phillips et al. (2017)
Successful
Model
Failed
Model
Positive Examples Negative Examples
32. QUESTION 1
• How many “nearest neighbor” chemicals are in the Boiling Point
OPERA model report for Bisphenol A
6 9 5 10
33. How to get there…
• Search for the chemical…
• Navigate to properties…
• Find the property of interest…
• Open the Predicted Properties and OPERA report…
34. QUESTION 1
• How many “nearest neighbor” chemicals are in the Boiling Point
OPERA model report for Bisphenol A
6 9 5 10
35. Different QS(X)R predictions
• What is the different between qsrr and qsar? How is DFT being
utilized in this field?
• QSPR: quantitative structure–property relationships
• QSAR: quantitative structure–activity relationships
• QSUR: quantitative structure-use relationships
• QSRR: quantitative structure-(Chromatographic) retention relationships
37. ToxCast and Tox21 bioactivity data for hazard
screening and prediction.
37
• ToxCast: more assays, fewer chemicals, EPA-driven
• Tox21: fewer assays, mostly 1536, driven by consortium
• All Tox21 data are analyzed by multiple partners
• Tox21 data is available analyzed in the ToxCast Data
Pipeline and other pipelines as well
EPA’s ToxCast program at a glance
Tox21 robot
41. ToxCast covers a lot of biology but not all
ToxCast is growing over time.
Invitrodb version 3.3 (released August 2020) contained 17 different assay sources, covering (at
least) 491 unique gene-related targets with 1600 unique assay endpoints.
41
Assay source Long name Truncated assay source description
Some rough notes on the biology
covered
ACEA ACEA Biosciences real-time, label-free, cell growth assay system based on a microelectronic impedance readout Endocrine (ER-induced proliferation)
APR Apredica CellCiphr High Content Imaging system Hepatic cells (HepG2)
ATG Attagene multiplexed pathway profiling platform
Nuclear receptor and stress response
profile
BSK Bioseek BioMAP system providing uniquely informative biological activity profiles in complex human primary co-culture systems Immune/inflammation responses
NVS Novascreen large diverse suite of cell-free binding and biochemical assays.
Receptor binding; transporter protein
binding; ion channels; enzyme inhibition;
many targets
OT Odyssey Thera novel protein:protein interaction assays using protein-fragment complementation technology Endocrine (ER and AR)
TOX21 Tox21/NCGC
Tox21 is an interagency agreement between the NIH, NTP, FDA and EPA. NIH Chemical Genomics Center (NCGC) is the primary screening facility
running ultra high-throughput screening assays across a large interagency-developed chemical library
Many – with many nuclear receptors
CEETOX Ceetox/OpAns HT-H295R assay Endocrine (steroidogenesis)
CLD CellzDirect
Formerly CellzDirect, this Contract Research Organization (CRO) is now part of the Invitrogen brand of Thermo Fisher providing cell-based in
vitro assay screening services using primary hepatocytes.
Liver (Phase I/Phase II/ Phase III
expression)
NHEERL_PADILLA NHEERL Padilla Lab
The Padilla laboratory at the EPA National Health and Environmental Effects Research Laboratory focuses on the development and screening of
zebrafish assays.
Zebrafish terata
NCCT NCCT Simmons Lab
The Simmons Lab at the EPA National Center for Computational Toxicology focuses on developing and implementing in vitro methods to identify
potential environmental toxicants.
Endocrine (thyroid - thyroperoxidase
inhibition)
TANGUAY Tanguay Lab The Tanguay Lab, based at the Oregon State University Sinnhuber Aquatic Research Laboratory, uses zebrafish as a systems toxicology model. Zebrafish terata/phenotypes
NHEERL_NIS
NHEERL Stoker &
Laws
The Stoker and Laws laboratories at the EPA National Health and Environmental Effects Research Laboratory work on the development and
implementation of high-throughput assays, particularly related to the sodium-iodide cotransporter (NIS).
Endocrine (thyroid - NIS inhibition)
UPITT
University of
Pittsburgh
The Johnston Lab at the University of Pittsburgh ran androgen receptor nuclear translocation assays under a Material Transfer Agreement (MTA)
for the ToxCast Phase 1, Phase 2, and E1K chemicals.
Endocrine (AR related)
47. Learning more about the assay
endpoints and biology
Download summary information here: https://www.epa.gov/chemical-research/exploring-toxcast-data-downloadable-data
Assay
endpoint
Assay
component
Assay
CEETOX_H295R
ESTRADIOL
ESTRADIOL_up
ESTRADIOL_dn
TESTOSTERONE
TESTOSTERONE_up
TESTOSTERONE_dn
https://comptox.epa.gov/dashboard/assay_endpoints/
Example assay annotation hierarchy
• Many assay endpoints mapped to a gene, if applicable
• Assay endpoints now cover >1000 unique gene targets in invitrodb 3.3
• Intended target family helps understand biological target(s)
• Apolipoprotein
• Apoptosis
• Background measurement
• Catalase
• Cell adhesion
• Cell cycle
• Cell morphology
• CYP
• Cytokine
• Deiodinase
• DNA binding
• Esterase
• Filaments
• GPCR
• Growth factor
• Histones
• Hydrolase
• Ion channel
• Kinase
• Ligase
• Lyase
• Malformation (zebrafish)
• Membrane protein
• Mitochondria
• Methyltransferase
• microRNA
• Mutagenicity response
• Nuclear receptor
• Oxidoreductase
• Phosphatase
• Protease/inhibitor
• Steroid hormone
• Transferase
• Transporter
48. What can be done with ToxCast data?
• (for example) Does this
substance have endocrine or
liver-mediated bioactivity?
• Is there support for one or
more adverse outcome
pathways based on these
data, or does the substance
appear “non-selective?”
48
• What is the relative priority of
this substance for additional
evaluation?
• Can a protective bioactivity-
based point-of-departure be
calculated?
Answering biological questions Answering risk-related questions
55. Cytotoxicity Threshold
55
This is the cytotoxicity threshold or
“burst” based on the method
described in Judson et al. 2016. It is
the lower bound on the estimate of
a cytotoxicity threshold.
56. What to make of the data
• Bisphenol A clearly has some in vitro nuclear receptor activity at
concentrations that may be below or near cytotoxicity.
• It has moderate ToxCast ER agonist and AR antagonist scores.
• The cytotoxicity threshold or “burst” seems to support selectivity of some
nuclear receptor responses.
• Diving a little deeper into the intended target family supports this analysis.
57. QUESTION 4
• How many assay endpoints are associated with the ACEA
vendor family of assays?
16 11 6 21
63. For Endocrine (AR and ER) better
to use summary models
Positive ToxCast ER pathway agonist
and ToxCast AR antagonist scores.
CERAPP = consensus ER QSAR (from 17 groups)
COMPARA = consensus AR QSAR
ToxCast Pathway Model AUC ER = full ER model (18 assays)
ToxCast Pathway Model AUC AR = full AR model (11 assays)
64. QUESTION 5
• How many bioactivity assays are associated with the ESR1
(estrogen receptor 1 [ Homo sapiens (human) ]) gene?
24 30 32 23
65. Where do we look for assay details?
• How do we search the 100s of genes mapped against assays
• Home page: Assay/Gene Search
66. QUESTION 5
• How many bioactivity assays are associated with the ESR1
(estrogen receptor 1 [ Homo sapiens (human) ]) gene?
24 30 32 23
67. A note on ToxCast versioning
• Data change: curve-fitting, addition of new data
• Models change: improvements, more data, etc.
• The CompTox Chemicals Dashboard release from July 2020 is
now using ToxCast invitrodb version 3.3:
https://doi.org/10.23645/epacomptox.6062479.v5
• All ToxCast data and endocrine models (CERAPP, COMPARA,
ER, AR, steroidogenesis) can currently be accessed from within
invitrodb.
• Data downloads for NCCT: https://www.epa.gov/chemical-
research/exploring-toxcast-data-downloadable-data
• We anticipate a new ToxCast release in 2021. 67
68. QUESTION 6
• How many Assay Endpoints are reported through the Dashboard?
1759 1659 1569 569
69. With each release, more assay endpoints and more
chemical x endpoint data are released
Invitrodb version 3.3 (released August 2020) contained 17 different assay sources, covering (at least) 491 unique gene-
related targets with 1600 unique assay endpoints. Varying amounts of data are available for 9949 unique substances.
These assay endpoints were notable additions in invitrodb version 3.3.
69
Assay source Long name Truncated assay source description Some rough notes on the biology covered
NCCT_MITO
NCCT (now Center
for Computational
Toxicology and
Exposure)
Mitochondrial
toxicity
Respirometric assay that measure mitochondrial function in HepG2 cells
Multiple assay endpoints to evaluate mitochondrial
function
https://doi.org/10.1093/toxsci/kfaa059.
NHEERL_MED
NHEERL Mid-
Continent Ecology
Division
The EPA Mid-Continent Ecology Division of the National Health and Environmental Effects
Research Laboratory screened the ToxCast Phase 1 chemical library for hDIO1 (deiodinase 1)
inhibition as part of an ecotoxicology effort.
Endocrine (thyroid – hDIO1,2,3 inhibition)
https://doi.org/10.1093/toxsci/kfy302
STM Stemina Stem cell-based metabolomic indicator of developmental toxicity for screening.
Developmental toxicity screening – multiple assay
endpoints
https://doi.org/10.1093/toxsci/kfaa014
LTEA
Life Tech Expression
Analysis
Gene expression measured in HepaRG cells following 48 hr exposure
Liver toxicity model via transcription factor regulated-
metabolism and markers of oxidative/cell stress;
multiple assay endpoints
70. QUESTION 6
• How many Assay Endpoints are reported through the Dashboard?
1759 1659 1569 569
74. QUESTION 7
• How many physicochemical properties can be predicted using
realtime predictions on the Dashboard?
18 9 27 19
75. How are the bioactivity data used???
In Vitro-In Vivo Extrapolation (IVIVE)
Translation of in vitro high throughput screening requires chemical-specific toxicokinetic models
Needed for anywhere from dozens to thousands of chemicals
Exposure in vitro bioactive
concentration
Toxicokinetic model:
Absorption
Distribution
Metabolism
Excretion
Internal
concentration
Iin vivo
TK data
Concentration
Response
In vitro Bioactivity
Assay
76. IVIVE-based reverse toxicokinetics
Reverse dosimetry can be leveraged in IVIVE to estimate the exposure that would
produce the plasma concentration corresponding to bioactivity
High-throughput toxicokinetic (HTTK) approaches make it possible to predict doses
corresponding to in vitro bioactivity for thousands of chemicals.
2012
A subset of the papers
describing the
development of a high-
throughput toxicokinetic
approach
2017
2017
2017
2014 2015
2019
2014
76
77. QUESTION 8
• How many HTTK related lists are available on the Dashboard?
0 6 3 1
78. QUESTION 8
• How many HTTK related lists are available on the Dashboard?
0 6 3 1
81. Definitions: Read-Across
• Known information on the property of a substance
(source) is used to make a prediction of the same
property for another substance (target) that is
considered “similar”
81
Source chemical Target chemical
Property
Reliable data
Missing data
Predicted to be harmful
Known to be harmful
Acute fish toxicity?
82. GenRA (Generalised Read-Across)
• Predicting toxicity as a similarity-weighted activity of nearest
neighbors based on chemistry and/or bioactivity descriptors
• Goal: to systematically evaluate read-across performance and
uncertainty using available data
• The approach enabled a performance baseline for read-across
predictions of toxicity effects within specific study outcomes to
be established
82
84. Read-across workflow in GenRA
84
Decision
Context
Screening level assessment
of hazard based on
toxicity effects from
ToxRefDB
Analogue
identification
Similarity context is based
on structural
characteristics
Data gap
analysis for
target and
source
analogues
Analogue
evaluation
Evaluate consistency and
concordance of
experimental data of
source analogues across
and between endpoints
Read-across
Similarity weighted
average – many to one
read-across
Uncertainty
assessment
Assess prediction and
uncertainty using AUC and
p value metrics
96. QUESTION 10
• How many chemicals are in the “National Health and Nutrition
Examination Survey” list
412 214 124 >500
97. Conclusions
• NAMs are increasingly accepted as the path forward to move
away from animal testing
• EPA-CCTE (previously NCCT) has been at the forefront of
NAMs development for over 15 years – in vitro bioactivity
measurements and modeling, exposure modeling, in vitro to in
vivo extrapolation, QSAR/QSUR/QSPR prediction etc.
• The Dashboard offers a path in to source relevant data and
models generated from our work
97
100. You want to know more…
• Lots of resources available
• Presentations: https://tinyurl.com/w5hqs55
• Communities of Practice Videos: https://rb.gy/qsbno1
• Manual: https://rb.gy/4fgydc
• Latest News: https://comptox.epa.gov/dashboard/news_info
100
101. Acknowledgments
• Contact: Williams.Antony@epa.gov
• Thanks to John Cowden, Katie
Paul-Friedman, Grace Patlewicz
and John Wambaugh for slides
• Feedback and follow-up is
welcomed! Your questions help
• The dashboard is based on the
efforts of many more team
members than us. Many
collaborators provide data also.
101
EPA’s Center for Computational Toxicology and Exposure
102. Questions from the Audience
• “Is there a unified database of CYP450 and drug transporter
interactions parameters (Km, Ki, EC50, IC50, etc.) with drugs
and environmental chemicals”
• Transportal: https://transportal.compbio.ucsf.edu/
103. Other databases of interest
• PDSP Ki https://pdsp.unc.edu/databases/kidb.php
• BindingDB https://www.bindingdb.org/
104.
105. 1. What is the best way to move beyond using many NAMs just for prioritization? Prioritizing
chemicals for what? Moving into animal experiments we are trying to eliminate using?
2. How can we use information from NAMs to support experiments using traditional in-vivo
bioassays?
3. I am quite interested in applying various NAMs endpoints in risk assessment, specifically how said
data could be used in toxicity factor derivation (e.g., the identification of a PODhec). Do we see the
data as an informant of MOA/key events or dose-response or both? How do we get the risk
assessment community to rely more upon these types of data? How do we demonstrate that these
methods are as good or better than traditional models that have been relied upon for toxicity factor
derivation up until this point? What new uncertainties need to be accounted for using this type of
data?
4. How to obtain sufficient toxicological data in short time consuming and how to interpret the data
based on figures and numbers and ask critical questions?? Thank you!
5. if you can have extra resources for NAMS explanations not presented - breakdown of what needs
in NAMS tables. For instance, physiochemical explanations to discount aspiration hazard not in
NAMs? Only modeling? In vitro studies? Etc.
Notas del editor
May also include a variety of new testing tools, such as “high-throughput screening” and “high-content methods” e.g. genomics, proteomics, metabolomics; as well as some “conventional” methods that aim to improve understanding of toxic effects, either through improving toxicokinetic or toxicodynamic knowledge for substances.
May also include a variety of new testing tools, such as “high-throughput screening” and “high-content methods” e.g. genomics, proteomics, metabolomics; as well as some “conventional” methods that aim to improve understanding of toxic effects, either through improving toxicokinetic or toxicodynamic knowledge for substances.
Definitions:
In silico – computationally based
In chemico – chemical reactions, generally abiotic
In vitro – taking place outside a living organism
In vivo – taking place inside a living organism
Toxicokinetics – how a substance gets into the body and what happens to it there
POD – point of departure – point on a dose-response curve that marks the no effect or low effect level
May also include a variety of new testing tools, such as “high-throughput screening” and “high-content methods” e.g. genomics, proteomics, metabolomics; as well as some “conventional” methods that aim to improve understanding of toxic effects, either through improving toxicokinetic or toxicodynamic knowledge for substances.
Definitions:
In silico – computationally based
In chemico – chemical reactions, generally abiotic
In vitro – taking place outside a living organism
In vivo – taking place inside a living organism
Toxicokinetics – how a substance gets into the body and what happens to it there
POD – point of departure – point on a dose-response curve that marks the no effect or low effect level