BOSC Review Draft

The New Chemicals Collaborative Research
Program: Modernizing the Process and
Bringing Innovative Science to Evaluate
New Chemicals Under TSCA

A Summary Report to the Board of Scientific Counselors (BOSC) on an integrative research plan within
the 2023-2026 Chemical Safety for Sustainability Strategic Research Action Plan

October 2022

Office of Research and Development
Office of Chemical Safety and Pollution Prevention
U.S. Environmental Protection Agency

1


-------
BOSC Review Draft

Contents

Executive Summary	3

Acronyms	4

Background	5

New Chemical Risk Assessment Challenges	5

Strategic Research Planning in ORD	9

New Chemicals Collaborative Research Program (NCCRP)	10

Problem and Vision Statement	11

Proposed NCCRP Research Areas	12

Proposed Research Relevant to the NCCRP	16

1.	Update and Refine Chemical Categories	16

A.	Chemical category modernization approach	17

B.	Expansion and application of systematic read-across	18

2.	Develop and Expand Databases Containing TSCA Chemical Information	20

A.	Chemical structure, physicochemical and environmental fate properties	22

B.	In vivo hazard data	24

C.	Exposure data	26

3.	Develop and Refine (Q)SAR and Predictive Models for Physicochemical Properties, Environmental
Fate/Transport, Hazard, Exposure, and Toxicokinetics	28

A.	Informatics platform for (Q)SAR development, implementation, and data management	29

B.	Exposure predictions	31

C.	Toxicokinetic predictions	32

4.	Explore Ways to Integrate and Apply In Vitro NAMs in New Chemical Assessments	33

A.	Analytical quality control of chemicals	36

B.	Screening for human health	36

C.	Screening for ecological health	38

D.	Screening for inhalation exposures	39

E.	Additional bioactivity data	39

5.	Develop a TSCA New Chemicals Decision Support Tool to Modernize the Process	40

A.	Implementing the International Uniform Chemical Information Database (IUCLID) in ORD	41

B.	Collaboration between ORD and OPPT on IUCLID data	41

C.	Developing proof-of-concept decision support tool for new chemicals	42

Conclusion	44

2


-------
BOSC Review Draft

References	45

Execut

The US EPA Office of Research and Development (ORD) has been developing and evaluating new
approach methodologies (NAMs) and decision support tools for toxicology and exposure to build a next
generation risk assessment toolbox. In a new joint effort referred to as the New Chemicals Collaborative
Research Program (NCCRP), the US EPA Office of Chemical Safety and Pollution Prevention (OCSPP) and
ORD are working together to bring innovative approaches to address the requirements of the Toxic
Substances Control Act (TSCA) for the review of new chemicals. TSCA requires EPA to review all new
chemical substances (i.e., those not yet in commerce) and, as amended in 2016, make one of several
affirmative determinations regarding risks to human health and the environment. Based on the
determination, EPA may need take further action to prevent unreasonable risks before manufacturing
for the chemical can commence. With hundreds of new chemical notices per year and limited hazard
and exposure information, addressing these statutory requirements with sound science, transparency,
and consistency, while meeting tight statutory deadlines for decisions, requires continued evolution of
scientific methods, approaches, and tools. Modernizing the new chemicals review process and bringing
innovative science to inform risk assessment and decision making will help overcome information gaps
and help the Agency meet its statutory requirements in a timely, effective, and efficient manner.

Under the NCCRP, ORD is working with the Office of Pollution Prevention and Toxics (OPPT)
within OCSPP to advance five key Research Areas: (1) updates and refinements to chemical analog and
category approaches; (2) development and expansion of databases containing TSCA chemical
information; (3) development and refinement of predictive models for physicochemical properties,
environmental fate/transport, hazard, exposure, and toxicokinetics; (4) integration and application of in
vitro NAMs; and (5) development of a TSCA new chemicals decision support tool that utilizes curated
data. Each of these five Research Areas represents translation and extension of computational
toxicology research that has been in development under the vision of the CompTox BluePrint (Thomas
et al., 2019) and the EPA NAM Work Plan (USEPA, 2021b), which together form a strategic roadmap for:
developing and integrating NAMs to fill information gaps; establishing scientific confidence of NAM
application to regulatory toxicology; and engaging with stakeholders. The NCCRP was announced in
February 2022 followed by a public meeting in April 2022 (USEPA, 2022a). ORD has aligned research
planning for the NCCRP with the ORD Chemical Safety for Sustainability (CSS) Strategic Research Action
Plan (StRAP) for 2023-2026, which has been reviewed by ORD management, representatives of OCSPP,
the Board of Scientific Counselors (BOSC) executive committee, and other stakeholders. Alignment with
the StRAP ensures that research within the NCCRP supports broader ORD objectives, including
coordination of NAM and interactive tool development, as well as coordination of resources. In this
report, details about the research proposed in the CSS StRAP that are relevant for the NCCRP will be
summarized and are the focus of this document and review by the BOSC.

Research within the scope of the NCCRP to address new chemical assessment is expected to
have broad applicability. As such, research will extend beyond this four year StRAP cycle and may
involve future collaborations with other relevant ORD research programs such as Health and
Environmental Risk Assessment, other federal institutions (e.g., the Division of Translational Toxicology,
formerly known as the National Toxicology Program, at the National Institute of Environmental Health
Sciences in the National Institutes of Health), and regulatory toxicology experts at other agencies such as
the European Chemicals Agency and Health Canada. Potential engagement with other ORD research
programs and external collaborators will leverage additional expertise and resources. These

3


-------
BOSC Review Draft

collaborations and the OPPT implementation details are beyond the scope of this report and are thus
not part of the BOSC review.

Acronyms

Acronym	Explanation

AOP	Adverse Outcome Pathway

BOSC	Board of Scientific Counselors

CBI	Confidential Business Information

CpDat	Chemicals and Products Database

CSS	Chemical Safety for Sustainability [a research program within ORD]

CvT	Concentration versus Time Database

DSSTox	Distributed Structure-Searchable Toxicity Database [for chemistry information]

ECOTOX	ECOTOXicology Knowledgebase

EPA	Environmental Protection Agency

HERA	Health and Environmental Risk Assessment [a research program within ORD]

HTTK	High-throughput Toxicokinetics

HTPP	High-throughput Phenotypic Profiling [also known as Cell Painting]

IATA	Integrated Approaches to Testing and Assessment

IUCLID	International Uniform Chemical Information Database

IVIVE	In Vitro to In Vivo Extrapolation

MIE	Molecular Initiating Event

MMDB	Multimedia Monitoring Database

NAM	New Approach Methodology

NCCs	New Chemical Categories [see d	> L	;

NCD

NCCRP

OCSPP

OHT

OPPT

ORD

PFAS

POD

(Q)SAR

(Q)SUR

St RAP

TEST/WebTEST

ToxRefDB

ToxValDB

TSCA

UVCBs

New Chemicals Division [in OPPT]

New Chemicals Collaborative Research Program

Office of Chemical Safety and Pollution Prevention (OCSPP)

Organisation for Economic Co-operation and Development [OECD] Harmonized Template

Office of Pollution Prevention and Toxics

Office of Research and Development

Per- and Polyfluoroalkyl Substances

Point-of-Departure

(Quantitative) Structure-Activity Relationship; some are more or less quantitative
Quantitative Structure-Use Relationship
Strategic Research Action Plan

Toxicity Estimation Software Tool/Web Toxicity Estimation Software Tool
Toxicity Reference Database
Toxicity Value Database
Toxic Substances Control Act

Chemical substances of Unknown or Variable Composition, Complex Reaction Products and
Biological Materials

4


-------
BOSC Review Draft

Background

The US EPA Office of Research and Development (ORD) and the Office of Chemical Safety and
Pollution Prevention (OCSPP) have been working collaboratively for many years to apply new approach
methodologies (NAMs) to regulatory toxicology needs. Most recently, ORD engaged with the Office of
Pollution Prevention and Toxics (OPPT) on the National PFAS Testing Strategy (USEPA, 2021a) and A
Proof of Concept Case Study Integrating Publicly Available Information to Screen Candidates for Chemical
Prioritization under the Toxic Substances Control Act (TSCA) (USEPA, 2021c). The New Chemicals
Collaborative Research Program (NCCRP) is an ambitious planned collaboration that will enable next
generation risk assessment while addressing OPPT's regulatory needs and bolstering ORD's efforts to
develop NAMs. More specifically, the NCCRP seeks to both rapidly modernize available approaches,
including decision support tools, for new chemicals evaluation and also impact the engineering of the
databases, models, and tools that ORD is building for multiple stakeholders to execute the vision of the
CompTox BluePrint (Thomas et al., 2019) and the EPA NAMs WorkPlan (USEPA, 2021b). If successful,
with each 4-year research cycle, the NCCRP will enable progress in ORD and OPPT toward meeting their
respective goals to advance chemical risk assessment. These goals include greater acceptance and
scientific confidence in NAMs applied within the NCCRP; greater understanding of the future needs of
NAM development; and decision support tools that provide consistent, but iteratively improving, access
to and integration of myriad data sources with chemical information, including data derived from NAMs.

In this section, the regulatory toxicology challenges posed by TSCA, how strategic research
planning to address these challenges is proceeding, and the launch of the proposed NCCRP will be
presented as background prior to discussion of the planned research.

New Chemical Risk Assessment Challenges

The regulatory toxicology challenges faced by OPPT guide both immediate and long-term goals

for the NCCRP. On June 22, 2016, TSCA was amended by the Frank R. Lautenberg Chemical Safety for the
21st Century Act.1 At the US EPA, OPPT, within OCSPP, is responsible for carrying out the mandates of
TSCA, including provisions requiring the review, determination of unreasonable risk, and subsequent
management of any identified risks associated with both existing (those already in the marketplace) and
new (those that manufacturers intend to bring to market) chemicals. OPPT's New Chemicals Division
(NCD) is responsible for the review of new chemicals prior to introduction of a new chemical into U.S.
commerce (via either import or domestic manufacturing). NCD received an average of 500 new chemical

1 The Frank R, Lautenberg Chemical Safety for the 21st Century Act [ US EPA

5


-------
BOSC Review Draft

notices per year since 2010 and a total of over 50,000 since 1979. Reference in this report to "new
chemical/' "submissions" or "notices," is generally meant to be inclusive of all potential notice types
under TSCA Section 5 requiring review, such as significant new use notices2, low volume and low release
and exposure exemption notices, and pre-manufacture notices (PMNs). Depending on the type of
notice, the statute or appropriate regulations generally require EPA to make determinations within 30 to
90 days of notice receipt. Details on the full review process can be found on EPA's website.3

A first key challenge posed by new chemical assessment is the dearth of information available,
as many new chemicals have little to no chemical-specific information available. To address data gaps
for both human and environmental hazards, and exposure, OPPT has led the world in the use of
(quantitative) structure-activity relationships ((Q)SARs) and other predictive models and tools coupled
with the use of category-based approaches.4 The methods, approaches, and tools developed over the
past four decades have been used to carry out tens of thousands of evaluations under TSCA. ORD plans
to augment the currently available (Q)SAR, read-across, and predictive approaches with new chemical
groupings, systemized read-across approaches, and (Q)SARs developed and evaluated using
internationally recognized and established validation principles (OECD, 2007; OECD, 2014).

A second major challenge has been implementing the changes in statutory requirements under
TSCA for new chemicals, which now mandates reviews and determinations for all new chemicals and
thus increased efforts and resources required in making determinations as well as documenting and
supporting those determinations. Prior to the 2016 amendments, EPA could evaluate whether to "drop"
a new chemical from further review and determined that for roughly 80% of annual new chemical
submissions a full determination or further detailed analyses were not necessary. A submitter could
then commence manufacture of the new chemical upon expiration of the review period without
restriction. In addition to requiring affirmative determinations for all new chemicals, the 2016
amendments introduced several new possible determinations that EPA could make, including that the
chemical is "not likely" to present an unreasonable risk or that the available information is "insufficient"
to permit a reasoned evaluation. In total, the statute now sets forth five possible determinations:

2	Significant new use notices may not necessarily pertain to new chemical substances but are nonetheless part of
the new chemicals program and are submitted under TSCA Section 5. See: https://www.epa.gov/reviewing-new-
chemicals-under-toxic-substances-control-act-tsca/filing-significant-new-use-notice

3	https://www.epa.gov/reviewing-new-chemicals-under-toxic-substances-control-act-tsca

4	Predictive Models and Tools for Assessing Chemicals under TSCA

6


-------
BOSC Review Draft

1.	The chemical substance or significant new use presents an unreasonable risk of injury to
human health or the environment;

2.	The information available is insufficient to permit a reasoned evaluation of the health
and environmental effects of the chemical substance or significant new use;

3.	In the absence of sufficient information to make an evaluation, the chemical substance
or significant new use may present an unreasonable risk of injury to health or the
environment;

4.	The chemical substance is or will be produced in substantial quantities and the
substance either enters or may reasonably be anticipated to enter the environment in
substantial quantities or there is or may be significant or substantial exposure to the
chemical; or,

5.	The chemical or significant new use is not likely to present an unreasonable risk of injury
to human health or the environment.5

Further, the 2016 amendments to TSCA explicitly require EPA to review new chemicals under
the "conditions of use" - a phrase defined in the law to include the circumstances under which the
chemical is, "intended, known or reasonably foreseen to be manufactured, processed, distributed in
commerce, used or disposed of." The identification of conditions of use - particularly those that are
"reasonably foreseen" - presents a unique challenge for the New Chemicals Program, given the data
submitted by the manufacturer regarding the intended use are often limited, let alone any future use.
NAMs developed or used by ORD and other stakeholders may have the potential added benefits of
addressing additional hazard data gaps, identification of conditions of use, and furnishing more
information for making the required determination.

OPPT has long used predictive models and other non-vertebrate methods for evaluating new
chemical submissions. The 2016 amendments to TSCA reinforced the need for more predictive models
and non-vertebrate methods via addition of Section 4(h) entitled, Reduction of Testing in Vertebrates,
which requires that: "The Administrator shall reduce and replace, to the extent practicable, scientifically
justified, and consistent with the policies of this title, the use of vertebrate animals in the testing of
chemical substances or mixtures under this title..." (Section 4(h)(1)). This subsection further requires
that prior to EPA making a request or adopting a requirement for testing using vertebrate animals, to
consider, as appropriate and to the extent practicable and scientifically justified, reasonably available

5 Section 2604(a)(3) at 15 USC Chapter 53

7


-------
BOSC Review Draft

existing information, including toxicity information, computational toxicology and bioinformatics, and
high-throughput screening methods and prediction tools (Section 4(h)(1)(A)).6 Section 4(h)(2)(A)
required OPPT to release a Strategic Plan in 20187 to promote the development and implementation of
alternative test methods and strategies. As suggested in Section 4(h), new chemical reviews need to
continue to incorporate new, innovative methods, approaches, and tools to maintain a modern and
efficient process.

The requirement to review and make an affirmative determination on each new chemical
submission for their conditions of use, the accompanying need to support each determination with a
robust assessment, the 90-day review requirement, and TSCA's direction regarding non-animal testing,
all underscore the need for updated approaches for new chemical assessments. Additionally,
incorporating additional NAMs could increase both efficiency and transparency.

A third major challenge area can be summarized as a substantial informatic need, in which
increased computational accessibility of data and modernized chemical information management is
required to efficiently perform assessments. Inherently, some TSCA information may be claimed as
confidential business information (CBI), and current public availability of existing chemical data to
inform computational approaches to chemical categories, read-across, and (Q)SAR development may be
limited. The development of human health and ecological risk assessments for new chemicals has relied
heavily on the limited information and data provided in a new chemical submission, often associated
with CBI claims. Providing additional public data and utilizing modernized tools will increase both the
transparency in decisions and the amount of information available to support new chemical
determinations. Since the 2016 amendments to TSCA, efforts have been underway in OPPT to make
information claimed as TSCA CBI publicly available where the Agency has determined that the
information is not entitled to confidential treatment.8 OPPT and ORD plan to work in tandem to increase

6	Section 4(h)(1)(B) contains further requirements, including "encouraging and facilitating— (i) the use of
scientifically valid test methods and strategies that reduce or replace the use of vertebrate animals while providing
information of equivalent or better scientific quality and relevance that will support regulatory decisions under this
title; (ii) the grouping of 2 or more chemical substances into scientifically appropriate categories in cases in which
testing of a chemical substance would provide scientifically valid and useful information on other chemical
substances in the category; and (iii) the formation of industry consortia to jointly conduct testing to avoid
unnecessary duplication of tests, provided that such consortia make all information from such testing available to
the Administrator."

7	See: https://www.epa.gov/assessing-and-managing-chemicals-under-tsca/strategjc-plan-reduce-use-vertebrate-

animals-chemical

8	Go to: Confidential Business Information under TSCA [ US EPA

8


-------
BOSC Review Draft

the computational accessibility of non-confidential information, including chemical-specific information
that ORD already curates and compiles into publicly available databases.

Bringing together existing chemical data for TSCA-relevant chemicals and NAM information will
require long-term work, extending beyond 2023-2026, to modernize access and utilization of
heterogeneous data coming from disparate sources. OPPT is working to modernize its digital chemical
information system that supports the entirety of the TSCA program. Although the effort to modernize its
chemical information system is internal to OPPT, the modernization effort is necessary for full
implementation of the tools developed under the NCCRP and may be coordinated with collaborative
research activities under the NCCRP. Within work relevant to the NCCRP, ORD plans to implement an
International Uniform Chemical Information Database (IUCLID)9 environment for existing curated
databases and public document sources. This will begin with curated in vivo hazard data but will extend
to as many data types as possible. Building an IUCLID compatible data environment within ORD and
OPPT contributes to a long-term Agency goal of using information from multiple sources (e.g., public and
internal) and supports development of a decision support tool for new chemical assessments that can
accelerate the pace of risk evaluations under TSCA.

Strategic Research Planning in ORD

Research within ORD is guided by the EPA Strategic Plan,10 which delineates clear goals for

Agency decisions and actions. Based this Plan, ORD delivers research to meet both short- and long-term
Agency needs, to inform Agency decisions, and to support the needs of tribal, state, and community
partners. ORD coordinates this research through four-year planning cycles within six National Research
Programs, including Chemical Safety for Sustainability (CSS), which includes research relevant to the
NCCRP.

Due to the high visibility of this cross-cutting research, technical perspectives from the ORD
federal advisory committee, the BOSC, are being sought. A draft of the CSS StRAP for FY23-26, also
known as StRAP4, was already presented to the BOSC executive committee for review,11 with an
emphasis on the Topic, Research Area,12 and Output level details, which follow a hierarchical order. The

9	See https://iuclid6,echa,europa,eu/project-iuclid~6

10	See: https://www.epa.gov/planandbudget/strategjcplan

11	https://www.epa.gov/svstem/files/documents/2022-04/epa-ord ess-fy23-26-d raft-strap 3-28~2022.pdf

12	Note that "Research Area" in the StRAP is not equivalent to "Research Area" in the NCCRP. Research Area in the
StRAP groups scientific expertise at a high level. Research Area in the NCCRP refers to more specific collections of
research to be performed and are described herein as Research Areas 1-5.

9


-------
BOSC Review Draft

broad Topics provide overarching research foci, followed by Research Areas which group the science
expertise and research that will be assembled to address partner needs. Outputs provide even more
detail about the results that will be achieved under each Research Area. Outputs are composed of
Products, which are the tangible deliverables of the National Research Program. In the report herein,
details about the relevant Outputs and narrative summaries of multiple Products will be provided to give
an overarching view of the coordinated research relevant to the NCCRP across CSS in StRAP4. This
review of the planned ORD research relevant to the NCCRP by the BOSC provides another opportunity to
obtain stakeholder perspectives. Previously, a public meeting announcing the NCCRP sought feedback
on the five research areas proposed (and outlined below in Figure 1 and Table 1). Future
implementation details within OPPT, based on progress and application of research relevant to the
NCCRP, will be presented to other federal advisory committees that provide advice on TSCA-relevant
work at the EPA.

Much, but not all, of the proposed research within ORD that is relevant to the NCCRP is
consolidated in CSS StRAP Output 408.4, Innovating Science to Support New Chemicals Evaluation. This
Output includes research that addresses the needs of EPA's programs and regions, states, tribes, and
external partners as well as identified cross-cutting research priorities for CSS: developing a tiered
testing strategy, building confidence in NAMs, increasing data availability and accessibility, and
contributing to decision support and translation. The innovative science required to address the risk
assessment of new chemicals necessitates coordination and work across the CSS portfolio, beyond CSS
408.4 (see Figure 2), and may involve future collaborations with other relevant ORD National Research
Programs such as the Health and Environmental Risk Assessment research program.

New Chemicals Collaborative Research Program (NCCRP)

The NCCRP will likely involve other federal institutions (e.g., the Division of Translational

Toxicology, formerly known as the National Toxicology Program, at the National Institute of
Environmental Health Sciences in the National Institutes of Health) as well as collaborations with other
regulatory entities such as Health Canada and the European Chemicals Agency to leverage the expertise
and resources of these entities to address TSCA-specific needs as well as to enhance broad applicability
of the research. The results of the effort are expected to increase the efficiency of new chemical
reviews, but more importantly bring innovative science to new chemicals assessments and decisions for
protecting human health and the environment using the authority under TSCA Section 5.

10


-------
BOSC Review Draft

ORD's Center for Computational Toxicology and Exposure (CCTE) and Center for Public Health
and Environmental Assessment (CPHEA) have been working closely with OPPT to develop this
overarching research plan and coordinate activities. Additionally, internal and external partners will be
consulted for input and research contributions. Previously, an overview of the NCCRP was released for
public comment (USEPA, 2022a), with comments received in a docket.13 While the focus of planned
research relevant to the NCCRP falls within the next StRAP4 (FY23-26), the collaboration needed to
support modernization and innovation for new chemicals assessment will likely extend beyond
completion of StRAP4.

Problem and Vision Statement

Refinement of and updates to methods, approaches, and tools used by OPPT to evaluate new

chemicals are critical to continuing to ensure the safety of new chemicals prior to their entrance into US

commerce and that decisions regarding the risk posed by new chemicals to human health and the

environment are supported by the best available science. Any changes should align with statutory

deadlines, be operational in a data poor environment, make effective use of new data sources and

approaches, and be transparent to the extent practicable (given that TSCA CBI may be used in the

development of these approaches). The vision of the NCCRP is to modernize the process for evaluating

new chemicals under TSCA by supporting the evolution of OPPT's use of new and existing methods,

approaches, and tools through the use of innovative science.

13 See: httpsi//www.epa.gov/reviewing-new-chemicals-under-toxic-substances-control-act-tsca/new-chemicals-
collaborative

11


-------
BOSC Review Draft

Proposed NCCRP Research Areas

The five proposed Research Areas are described and summarized in Figure 1 and Table 1.

Figure 1. Interconnectivity of the NCCRP Research Areas.

These five Research Areas are interconnected efforts to ultimately integrate NAMs and computationally
accessible data into a decision support tool that can be iteratively improved to support new chemical
assessments. Research areas 1, 2, 3, and 4, bounded by a dashed rectangle, are all interrelated;
computationally accessible data in Research Area 2 feeds into Research Areas 1 and 3, and informs data
gaps to be addressed in Research Area 4. Research Areas 1, 3, and 4 supply data back to the database
environment. Together, Research Areas 1-4 supply information to be used in Research Area 5
(development of a decision support tool). Research Areas: 1 = Update and refine chemical categories; 2
= Develop and expand databases containing TSCA chemical information; 3 = Develop and refine (Q)SAR
and predictive models for physicochemical properties, environmental fate/transport, hazard, exposure,
and toxicokinetics; 4 = Explore ways to integrate and apply in vitro NAMs in new chemical assessments;
5 = Develop a decision support tool to modernize the process.

Develop and expand databases

Machine readable data
Manual curation
Literature mining

1 Chemical categories
and analogs

3 (Q)SARs and
predictive models

In vitro assay
data and models

Decision support

Collate

Display M

Annotate n

Report

information

information

and Decide



12


-------
BOSC Review Draft

Table 1. Proposed NCCRP Research Areas.

Each Research Area is addressing key problems, and through applied research approaches, will yield one
or more outcomes that have the potential to impact new chemical decisions in OPPT.

Research Area Problem	Approach

Expected Outcome(s)

1

Update and

Currently 56

Systematically define chemical

This will increase the efficiency



Refine Chemical

TSCA categories,

categories and analogs for read-

of new chemical reviews and



Categories

last updated

across using structural (and other)

promote the use of the best





2010

boundaries; physical-chemical

available data to protect







properties; structural alerts for

human health and the







hazard, fate, exposure, and/or

environment.







functional uses; existing hazard









data; and/or, in vitro mechanistic









and toxicokinetic data from NAMs



2

Develop and

Existing TSCA

Extract and curate available TSCA

The TSCA CBI information will



Expand

information is

CBI study information

be combined with publicly



Databases

not



available sources to expand



Containing TSCA

computationally

Continue extraction and curation

the amount of information



Chemical

accessible or

of physical-chemical property,

available, enhancing chemical



Information

easily searchable

environmental fate, hazard, and

reviews and enabling efficient







exposure information (non-CBI) in

sharing of chemical







ORD databases

information across EPA.









Safeguards for CBI will be







Map information in ORD

maintained as appropriate in







databases to standardized

this process.







reporting templates and store in









an International Uniform Chemical









Information Database (IUCLID)



3

Develop and

Currently used

Develop and update (Q)SAR and

Updated models that reflect



Refine (Q)SAR

models are not

predictive models using existing

the best available science,



and Predictive

always publicly

data and curated data from

increased transparency, and a



Models for

accessible, easy

Research Area 2

process for updating these



Physical-

to update with



models as science allows.



Chemical

additional

Evaluate models to determine the





Properties,

chemicals, or the

best suite for use by OPPT for





Environmental

best performing

regulatory purposes





Fate/Transport,

for all







Hazard,

chemistries







Exposure, and









Toxicokinetics







13


-------
BOSC Review Draft

4 Explore Ways to Reduction in the

Integrate and
Apply In Vitro
NAMs in New
Chemical
Assessments

use of vertebrate
animals in
accordance with
TSCA Section
4(h)

Many new
chemical
submissions are
data poor

Develop and evaluate a suite of in
vitro NAMs for informing new
chemical evaluations

Use mechanistic and toxicokinetic
in vitro NAMs to inform and refine
chemical categories in Research
Area 1

A suite of in vitro NAMs that
could be used by external
stakeholders for testing and
data submissions under TSCA
as well as informing and
expanding new chemical
categories

Amended TSCA

requires

affirmative

determination

regarding

unreasonable

risk

Develop a TSCA
New Chemicals
Decision Support
Tool to

Modernize the
Process

Searching,
collating, and
integrating data
for new chemical
assessments is
inefficient and
costly

Build proof of concept software
workflow that integrates all data
streams in a new chemical risk
decision context

A decision support tool that
will efficiently integrate all the
data streams (e.g., chemistry,
fate, exposures, hazards) into
a final risk assessment and
transparently document the
decisions and assumptions
made. This will facilitate the
new chemicals program
tracking decisions over time
and evaluating consistency
within and across chemistries.

14


-------
BOSC Review Draft

Figure 2. NCCRP Research Areas rely on CSS research.

The five NCCRP Research Areas cut across several CSS StRAP Outputs previously reviewed by the BOSC,
ORD, and stakeholders. Blue boxes represent overlap of NCCRP research activities with CSS StRAP
Outputs.

CSS Research
Area

High-
throughput
toxicology (HTT)

Rapid Exposure
and Dosimetry
(RED)

Ecotoxicological
Assessment and
Modeling
(ETAM)

Chemical
Characterization
and Informatics
(CCI)

Integration,
Translation, and
Knowledge
Delivery (ITK)

Output

401.1

1 Update
and refine
chemical
categories

2 Develop
and expand
databases

3 Develop
and refine
QSAR and
predictive
models

4 Explore
ways to
integrate
and apply
NAMs

5 Develop a
decision
support
tool

408.4

15


-------
BOSC Review Draft

Proposed Research Relevant to the NCCRP

1. Update and Refine Chemical Categories

As TSCA new chemical notices are typically data poor, OPPT has historically relied heavily on the
use of chemical categories14 and read-across15 as methods to fill data gaps, particularly for hazard
characterization. OPPT currently uses the 2010 version of the New Chemicals Program under TSCA
Chemical Categories document which identifies 56 chemical categories16 based on chemical class,
referred to herein as new chemical categories (NCCs). When OPPT evaluates a new chemical,
determining if it belongs in an existing NCC is important for evaluating human health or environmental
effects.

In Research Area 1, ORD and OPPT are proposing to develop a systematic, transparent, and
reproducible approach for modernizing both chemical categories and read-across methods. Research
will identify scientific information to support development or refinement of chemical categories and
read-across methods, such as: structural (and other) boundaries; physicochemical properties; structural
alerts for hazard, fate, exposure, and/or functional uses; mechanistic and toxicokinetic data from NAMs;
and/or, existing hazard data. The new approach will document the data used to inform chemical
categories as well as the basis of any similarity or read-across applications in a systematic manner.

The proposed approach will increase the efficiency of new chemical reviews and promote the
use of the best available data to protect human health and the environment. Further, application of a
chemical category approach itself should result in greater confidence in inferences made for a given

14	These categories are used for analysis and risk management of individual new chemicals, and thus do not
implicate Section 26(c) of TSCA, which allows EPA to take action with respect to a category of chemical substances.
Nonetheless, the new chemicals categories use similar principles to categories under Section 26(c), which may be
applied to, "a group of chemical substances the members of which are similar in molecular structure, in physical,
chemical, or biological properties, in use, or in mode of entrance into the human body or into the environment, or
the members of which are in some other way suitable for classification as such for purposes of [TSCA.]" The
Organization for Economic Cooperation and Development (OECD) defines a category as "(C)hemicals whose
physicochemical, toxicological and ecotoxicological properties are likely to be similar or follow a regular pattern as
a result of structural similarity may be considered as a group, or 'category'..." (p.11 in OECD, 2017).

15	Read-across is defined as a data gap filling technique that relies on an analog or category approach, with analogs
or categories defined on the basis of similarity of structure, properties, or other information. To "read across" is to
apply data from a tested chemical for a particular property or effect to similar untested chemicals. See
https://dx.doi.org/10.14573/altex.1410071 and https://dx.doi.Org/10.1016/i.vrtph.2016.05.008 for further
discussion.

16	See Chemical Categories Used to Review New Chemicals under TSCA

16


-------
BOSC Review Draft

target chemical (OECD, 2014), assuming that the categories applied are robust, which depends on
category size (number of members) and the amount of data available for each category member. This
research builds upon ongoing research and further motivates applied cheminformatic research within
ORD to support CSS goals. The research covers several CSS StRAP Outputs listed in Table 2.

Table 2. StRAP Outputs Relevant to Update and Refine Chemical Categories





CSS.407.3

Develop new and improve existing structure activity relationship models
to support risk assessment

CSS.407.4

Advancing chemical categorization approaches for aiding the
interpretation and prediction of bioassay and toxicity outcomes

CSS.407.5

Advancing the use of structural, mechanistic, and toxicokinetic data to
support categorization and classification of Per- and Polyfluoroalkyl
substances (PFAS)

CSS.408.4

Innovating science to support new chemicals evaluation

A. Chemical category modernization approach

The 56 existing NCCs are characterized largely by structural features and in some cases by
physicochemical properties. The key goals of collaborative research in this area are to implement the
chemical categories in a transparent and reproducible manner that would permit updates with new
information, such as additional structure descriptors, physicochemical data, or NAM data. Further,
planned research will investigate to what extent new categories are needed to capture substances in the
TSCA active inventory that could not be readily assigned to one of the 56 existing NCCs. This research
will bridge between the current NCCs and development of an easily updated approach to chemical
grouping.

First, the chemical structure information built into the current NCCs will be turned into a machine-
readable format, such as system arbitrary target specification (referred to as SMARTS), to enable
substructure searching and mapping to other types of structural descriptors, such as ToxPrints (Yang et
al., 2015). This research will enable computational approaches to chemical grouping based on one or
more types of structural descriptor(s) as well as other pertinent information. The TSCA non-confidential
active chemical inventory will be profiled using the newly codified NCCs to assign them into their
respective categories. Chemical categories may be developed by a combination of one or more of the
following: use of more structural descriptors, physicochemical properties, predicted metabolism, in vitro
mechanistic and toxicokinetic, and/or in vivo toxicity data, pending resources and available data. Finally,
with research completed to better understand the chemical structure space encompassed within the

17


-------
BOSC Review Draft

TSCA non-confidential active chemical inventory, research will evaluate to what extent the chemicals on
this inventory fall within the applicability domain for existing ORD (Q)SAR models such as the Toxicity
Estimation Software Tool (TEST) (USEPA, 2020b), with an online version referred to as WebTESTl.O, or
other structural alert schemes (either existing or in development) to better characterize limitations in
the ability of those models to make robust and reliable predictions. This will help target further data
curation efforts for chemistry information in Research Area 2 aimed at trying to increase the
applicability domain of structure alerts and models. The insights gained will help tailor the combination
of NCCs and models as a proof-of-principle scheme that is most informative for in silico evaluation of the
TSCA active inventory.

B. Expansion and application of systematic read-across

Generalized read-across (GenRA) (Helman et al., 2019a; Helman et al., 2018; Helman et al., 2019b;
Shah et al., 2016; Shah et al., 2021), is a systematic, data-driven read-across approach developed by
ORD. GenRA has been implemented as part of a read-across workflow within a web application initiated
via chemical search or chemical structure drawing in the CompTox Chemicals Dashboard17 and also as a
Python package, genra-py, to facilitate batch processing with user-specific datasets. Though the GenRA
approach has been applied to systematically evaluate in vivo toxicity datasets represented by potency
values or binary hazard outcomes, the GenRA web application is currently structured to make binary
(positive or negative) in vivo toxicity estimates based only on data curated into the Toxicity Reference
Database (ToxRefDB). Local neighborhoods characterized by different fingerprints based on chemistry
and/or bioactivity information, e.g., circular Morgan fingerprints (Rogers and Hahn, 2010), ToxCast
bioactivity fingerprints based on positive or negative assay responses, or a hybrid combination of both,
can be used to identify candidate source analogs from which a GenRA prediction is derived for the
toxicity outcomes of interest. GenRA is a tool that could be potentially implemented by the OPPT NCD in
the near-term, and such translation efforts will be part of the NCCRP. GenRA can also be enhanced to
better meet the needs of the NCCRP over the course of StRAP4, benefitting not only new chemical
assessment, but also other applications of GenRA.

Key planned research to enhance GenRA capabilities, and also meet OPPT needs for transparent and
reproducible read-across, will proceed in a number of areas, including: evaluating the impact of hybrid
features on GenRA performance; extending similarity contexts to additional types of bioactivity data;

17 https://comptox.epa.gov/dashboard/

18


-------
BOSC Review Draft

evaluating the contribution of metabolism data to inform analog identification and evaluation; and,
additional case studies to build confidence in the use of GenRA.

Within the current GenRA web application, a hybrid fingerprint can be constructed to search for
candidate source analogs. This hybrid can currently be created by using up to three different fingerprint
types with associated percentage weightings, e.g. 60% contribution from Morgan fingerprints versus
40% ToxCast bioactivity fingerprints. However, guidance as to what might be an optimal weight to use
for each fingerprint type, and the extent to which this differs depending on the type of chemical (e.g.,
structural class, functional groups) or toxicity endpoints being predicted (e.g., liver toxicity versus kidney
effects), has not been systematically evaluated. Such an analysis will characterize the relative
contribution that different weighting schemes may play in predicting toxicity outcomes.

Additional information to define target to analog similarity, such as in vitro bioactivity, could be very
informative for analog selection. An in vitro NAM such as high-throughput phenotypic profiling (HTPP) is
broad in biological coverage, high-throughput, and multi-dimensional, and may be useful in
understanding the bioactivity fingerprints of chemicals that share common biological targets (Nyffeler et
al., 2022; Nyffeler et al., 2020; Willis et al., 2020). Other multi-dimensional assay suites, such as a safety
pharmacology panel (Bowes et al., 2012; letswaart et al., 2020; Smit et al., 2021; Valentin et al., 2018),
may also provide valuable mechanistic fingerprint data to evaluate the similarity of potential analogs for
a target chemical. Additional research within GenRA aims to quantify the potential contribution of these
bioactivity data in inferring toxicity in read-across applications.

Analog identification would be further informed by understanding related metabolites of a target.
Initial planned research includes in silico predictions of liver-generated metabolites using a selection of
prediction tools (e.g., BioTransformer, OECD Toolbox) across a large and diverse chemical data set (e.g.,
ToxCast chemical library). An accessible database of these metabolite predictions provides a foundation
for investigation of chemical structural similarity and common metabolic pathways to better inform
chemical categories and analog identification. Chemical similarity in related metabolite production,
whether that be by virtue of the similarity in transformation profile, the sequence of transformations, or
the structural similarity in the predicted metabolites themselves, will be determined. A possible
extension to GenRA will aim to complement ongoing work in other areas of CSS to collect experimental
metabolism information.

19


-------
BOSC Review Draft

Finally, additional case studies will bolster confidence in the application of GenRA in regulatory
toxicology. GenRA performance on in vivo toxicity datasets has been systematically evaluated using
standard performance metrics, such as a coefficient of determination, to understand the goodness of fit
for GenRA predictions. However, a comparison with expert driven read-across cases has not been
performed primarily because a database of expert-driven read-across selections is not available. A
concerted effort will be made to identify read-across case studies either reported in the literature or
under the auspices of other EPA or OECD activities and extract relevant information, including: the
strategy taken to identify candidate source analogs, the rationale for selecting source analogs, the
toxicity endpoint being predicted, the underlying toxicity data for the source analogs, and the data gap
filling technique used.

2. Develop and Expand Databases Containing TSCA Chemical Information

In addition to the information submitted for a new chemical, information on other TSCA-
relevant chemicals may be found in a wide variety of public sources as well as in legacy OPPT TSCA files
(which may include TSCA CBI). However, many of the public sources as well as the TSCA data are not in a
digital form that can be efficiently searched, analyzed, and used to develop and refine (Q)SAR models,
inform the refinement and development of chemical categories, and provide data for analogs in read-
across evaluations of new chemicals. ORD is proposing to continue expanding existing ORD databases
and curation efforts to structure data including physicochemical and environmental fate properties
(ChemProp); household product chemical composition and function (CPDat); multimedia monitoring
data (MMDB); ecological hazard (ECOTOX); human health hazard (ToxVal, ToxRefDB); and toxicokinetics
(CvT, HTTK). The information in the ORD databases will be mapped to available standardized reporting
templates (starting with hazard data), stored in IUCLID as appropriate (see Research Area 5), and made
publicly available. Literature mining tools for information retrieval and extraction will also be refined
and further developed to attempt to rapidly screen the open literature and gray literature, i.e.
information not available from commercial publishers, for relevant information on chemicals and
associated analogs, with publication of the tools and approaches employed.

In addition, OPPT will plan to identify, extract, curate and catalog available data on chemistry,
hazard, fate, and exposure from different TSCA databases and holdings (which may include TSCA CBI).
This will include digitizing existing physical records (largely paper and some microfiche) to capture all
relevant TSCA information for a given chemical substance. Available TSCA information will be combined

20


-------
BOSC Review Draft

with publicly available sources, such as information from ORD databases, to expand the amount of
information available, thereby enhancing chemical reviews and enabling efficient sharing of chemical
information across EPA. Safeguards for CBI will be maintained as appropriate in this process. This
initiative to digitize TSCA information contributes to a long-term goal of maintaining and utilizing fully
computationally accessible data within OPPT. Based on the size of this task, ongoing work within OPPT
to achieve this extends beyond the initial collaborative research plan described here.

The proposed research in Research Area 2 and Research Area 5 will result in EPA having
increased interoperability between relational databases containing TSCA-relevant physicochemical
properties, environmental fate/transport, hazard, and exposure information to ensure the efficient
searching of existing chemical information. Additional curation efforts will expand available chemical
information for developing chemical categories, (Q)SAR, and other predictive models and will enable
efficient sharing of chemical information within EPA. This research builds upon ongoing data curation
activities and further motivates curation and database engineering tasks to support TSCA new chemical
assessments and collaborations with internal and external stakeholders that utilize harmonized data
formats, particularly for hazard data (Table 3).

21


-------
BOSC Review Draft

Table 3. StRAP Outputs Relevant to Data Curation.

Data type curated

Relevant StRAP4
Output

Relevant StRAP4 Output Title

Chemistry and

CSS.407.1

Generate and curate data relevant to chemical

properties



substances, structures, samples, and properties

Chemistry and

CSS.407.3

Develop new and improve existing structure

properties



activity relationship models to support risk
assessment

Chemistry and

CSS.407.5

Advancing the use of structural, mechanistic,

properties



and toxicokinetic data to support categorization
and classification of Per- and Polyfluoroalkyl
substances (PFAS)

In vivo hazard in human

CSS.401.2

Provide structured and computationally

health relevant models



accessible data to support tiered toxicity testing



CSS.408.4

Innovating science to support new chemicals
evaluation

In vivo hazard in

CSS.406.3

Identify, assemble, and curate toxicity data for

ecologically relevant



ecologically relevant species for risk assessment

species



(ECOTOX)

Monitoring and release

CSS.402.1

Collect and curate exposure-relevant data

data to support





exposure assessment





and modeling





Toxicokinetic data

CSS.402.2

High-throughput toxicokinetic (HTTK) tools to
support in vitro to in vivo extrapolation

A. Chemical structure, physicochemical and environmental fate properties

Expanded curation of chemical identity, physicochemical, and environmental fate properties make
more of the chemical landscape accessible for chemical category formation, read-across, and predictive
modeling. As more chemicals are added to the TSCA active nonconfidential inventory, or chemistries
with limited available information are identified, more structure, physicochemical, and environmental
fate property data curation is needed to support decision making.

1. Chemical identity and structure (Distributed structure-searchable toxicity, DSSTox): Each time
a new chemical submission is reviewed, an initial step may include connecting the submitted
chemical structure to existing information or to existing information for analogs (on the basis of
attributes such as structure). Thus, chemical identity and structure curation that expands along
with expansions of the TSCA chemical inventory will support new chemical evaluation. Ongoing
work in ORD to expand and update the DSSTox database (Grulke et al., 2019) informs the
CompTox Chemicals Dashboard (Williams et al., 2017) and internal chem- and bio-informatic

22


-------
BOSC Review Draft

databases and models that may be relevant for application within the NCCRP. For instance,
(Q)SAR development and surfacing of curated hazard and exposure information all rely upon
accurately curated chemistry information. DSSTox chemical information spans over 1.2 million
unique substances (as of July 2022) with definitions of chemical structure and with curated
identifier linkages. Accurate structure-data linkages, in turn, provide the quality foundation for
chemical screening projects (such as ToxCast and high-throughput screening efforts), read-
across, non-targeted analysis, and structure-based modeling efforts across CSS research
programs, including models useful to the NCCRP. As the TSCA active inventory of chemicals
grows each year, expansion of the DSSTox database to include these chemicals as well as
existing and emerging chemicals of interest for modeling applications is essential. This DSSTox
expansion supports the evolving needs of the Agency and providing programmatic access to any
data that can be linked to a DSSTox identifier. New chemical submissions under TSCA may be for
defined or complex mixtures, and chemistry curation can provide solutions for better linking
appropriate data to these mixtures to facilitate read-across or other downstream predictions.
ORD efforts to curate special chemistries relevant to TSCA, such as PFAS, require manual
curation using Markush-type structure representations (as exemplified by the PFAS Category
list: https://comptox.epa.gov/dashboard/chemical-lists/EPAPFASCAT) or creation of linkages to
defined structural components, in order to enable connection of these chemical structures to
the web-accessible compendium of structure-linked chemical knowledge. In addition to
providing linkage of families of chemicals that have similar structural features, chemistry
curation of Markush representations helps address challenges in assigning defined structures to
complex mixtures, including UVCBs (i.e., chemical substances of Unknown or Variable
Composition, Complex Reaction Products and Biological Materials). Indeed, UVCBs constitute a
significant percentage of TSCA (i.e., over 40% of the non-confidential TSCA chemical inventory
list on the Dashboard, https://comptox.epa.gov/dashboard/chemical~
lists/TSCA ACTIVE NCTI 0320, cannot be assigned defined structures). These chemistry
curation efforts provide a foundational source of chemistry information to be used in research
and workflow applications for the NCCRP.

2. Physicochemical, fate, and toxicity properties: Within ORD, the development and evaluation of
(Q)SAR models for physicochemical, fate, and toxicity properties have relied on curated data
from external sources. Making forward predictions of these properties for new chemical
submissions may improve with more data from existing TSCA-relevant chemicals. In this

23


-------
BOSC Review Draft

proposed research, additional property data relevant to training and test sets for (Q)SARs
developed for or applied to the NCCRP will be curated. For example, (Q)SAR prediction models
from the Toxicity Estimation Software Tool (TEST) were built on datasets compiled before 2012
and have training data that are distinct and generally smaller than those used to build Open
Structure-Activity/Property Relationship Application (OPERA) models (Mansouri et al., 2018;
Mansouri et al., 2016). Recent efforts greatly expanded available property data, including data
for PFAS. As more TSCA-relevant chemicals are added to DSSTox, or as gaps in chemical category
coverage are identified through Research Area 1, additional physicochemical, fate, and toxicity
property data will be prioritized for curation.

B. In vivo hazard data

Expanded curation of in vivo hazard data will be performed to make more information available on
potential analogs for target new chemical submissions under TSCA. For example, in silico approaches
such as GenRA rely upon quality curation and computationally accessible in vivo hazard data in
ToxRefDB. In vivo hazard data curation will be prioritized to capture chemistries of interest based on
learnings from chemical categories in Research Area 1, information from OPPT, and available data.

These curated data can be described as human health hazard relevant (for ToxRefDB and/or ToxValDB)
or ecologically relevant (for ECOTOX). These data curation efforts are described in more detail below.

1. Human health hazard data (ToxRefDB and ToxValDB): In ORD, legacy in vivo human health
hazard information is stored in two different relational databases: the Toxicity Reference
Database (ToxRefDB) and the Toxicity Value Database (ToxValDB). These two databases
continue to be expanded through manual, user interface-driven curation workflow and scraping
of web-accessible data, respectively. As an example, TSCA-relevant chemical data may be added
in ToxRefDB and/or ToxValDB via curation of full text source documents and aggregation of
summarized data available on the public website, ChemView.18 More hazard information,
particularly from guideline studies or guideline-like studies, will be added based on identification
of source documents or data for TSCA-relevant chemicals and resources available for curation.
Expanding the chemical and study coverage for these databases for TSCA-relevant chemicals will
improve in silico approaches, including (Q)SAR and read-across, for new chemical evaluation.

18 https://chemview.epa.gov/chemview/

24


-------
BOSC Review Draft

a.	ToxRefDB (Watford et al., 2019) contains in vivo study data from over 5900 guideline or
guideline-like studies for over 1100 chemicals. This is largely comprised of curated
animal study data from repeat dose studies conducted according to Health Effects Series
870 guidelines, and many of these studies (over 3,000 of them) come from registrant-
submitted toxicity studies known as data evaluation records (DERs) from the U.S. EPA's
Office of Pesticide Programs (OPP). While employing a controlled vocabulary for
enhanced data quality, ToxRefDB serves as a resource for study design, quantitative
dose response, and endpoint testing status information given guideline specifications.
To enable high quality and consistent extractions, the document extraction is performed
by 1-2 study curators with manager review within the Data Collection Tool (DCT).
ToxRefDB is summarized with calculated point-of-departure values at the chemical and
study level for inclusion in the summary-level database, ToxValDB.

b.	ToxValDB includes data on thousands of chemicals from tens of thousands of records,
with an emphasis on quantitative estimates of relevant points-of-departure from in vivo
toxicology studies, such as no- and low-observable adverse effect levels, screening
levels, reference doses, tolerable daily intake, etc. The source data originates from
multiple public datasets, databases (i.e., with data already digitized), and the open
literature. Each dataset is reshaped to a standard source data format and then all source
data streams are integrated into the main ToxValDB database. Data is reviewed for
quality within the source data tables. In addition to the main in vivo quantitative data,
ToxValDB also contains data on cancer slope factors, genotoxicity assays, and acute
toxicity information (e.g., skin and eye irritation and skin sensitization). All data is
surfaced on the CompTox Chemicals Dashboard.19

2. Ecologically-relevant hazard data (ECOTOXicology Knowledgebase, ECOTOX): ECOTOX (Olker et
al., 2022) is a comprehensive, publicly available knowledgebase providing single chemical
environmental toxicity data on aquatic life, terrestrial plants, and wildlife. Hazard data are
extracted and added to ECOTOX quarterly and improvement and expansion of the ECOTOX
controlled vocabulary is ongoing. Addition of data to ECOTOX may include TSCA-relevant
chemicals, as well as emerging contaminants, chemicals associated with assessment of
Endangered Species Act concerns, and/or chemicals of interest to the Office of Pesticide

19 CompTox Chemicals Dashboard: httpsi//comptox, epa.gov/dashboard/

25


-------
BOSC Review Draft

Programs. Additional curation of TSCA-relevant chemicals will enhance opportunities for read-
across and augmentation of training data in (Q)SAR models for aquatic toxicity. ECOTOX has its
own user interface20 and is also surfaced on the CompTox Chemicals Dashboard.

C. Exposure data

Expanding curation of the use context for chemicals supports development of quantitative structure
use relationships that may be important for predicting potential uses of new chemicals. Additional
curation of chemical occurrence in environmental matrices can inform predictions of chemical release to
environmental compartments for data-poor chemicals. And finally, curated in vivo toxicokinetic data can
support evaluations of confidence in high-throughput toxicokinetic (HTTK) modeling and curated
intrinsic clearance and plasma protein binding data can further inform generic toxicokinetic modeling
approaches applied for in vitro to in vivo extrapolation (IVIVE).

1. Chemicals and Products Database (CPDat): How a chemical is used in consumer, occupational,
or industrial context(s) determines relevant exposure pathways for risk evaluation. As a result,
exposure research relies on well-documented and accessible datasets of chemical use
information curated from publicly available sources. These curated chemical use data can
support key needs of new chemical assessments under TSCA with respect to development and
refinement of systematic approaches for predicting potential conditions of use for submitted
substances (e.g., via the development of Quantitative Structure Use Relationship, or (Q)SUR,
models) (Phillips et al., 2017); estimating exposure via poorly characterized release scenarios;
characterizing variability in population exposures; and, addressing data-poor chemicals by
creating generic exposure scenarios by use. ORD developed a data management and curation
application, Factotum, to facilitate the rapid collection and distribution of high-quality chemical
and exposure-related data from public documents via curation, quality assurance, visualization,
and data delivery tools, which are released as the Chemicals and Products Database (Dionisio et
al., 2018; USEPA, 2020a). Factotum has also enabled manual and machine-learning based
curation of data to new or updated harmonized chemical use tags, product categories, and
OECD function categories. Additional planned curation efforts will augment consumer product
information curation with data on the occurrence of data-poor chemistries, including PFAS and

20ECOTOX user interface: httpsi//cfpub.epa.gov/ecotox/

26


-------
BOSC Review Draft

UVCBs, in products within the home by analyzing data from measures of presence and
emissivity of these chemicals from household products and articles. The corpus of curated
chemical use information will help fill gaps in inferring exposure scenarios for new chemicals.

2.	Environmental occurrence (Multimedia Monitoring Database, MMDB): Chemical monitoring
data are the gold-standard for exposure data in risk assessment, and curation of additional
chemical monitoring data would inform predictive models of potential environmental
exposures. A recent OPPT and ORD collaboration produced a machine-readable database,
MMDB, which comprises chemical measurements in environmental media collected and curated
from publicly available government databases and reports as well as additional media
occurrence data (in consumer articles, indoor air and dust, and biological media) curated from
scientific literature (Isaacs et al., 2022; USEPA, 2022b). MMDB requires standardization and
aggregation of monitoring data for each media to harmonized units, where possible. Planned
curation work potentially includes, but is not limited to, quantitative or qualitative non-targeted
analysis data from EPA or other studies; new data on chemicals measured in biosolids; PCB
concentrations in consumer products; residential exposure data from EPA field studies; and
monitoring data for PFAS compounds obtained from extracted reports or other

sources. Currently, these data are considered public, but are generally inaccessible. A number of
these regulatory monitoring data streams are not currently in MMDB and will provide a unique
and valuable contribution to the database. In collaboration with the Center for Environmental
Solutions and Emergency Response (CESER), this product will also produce a production-quality
database (tentatively called StEWIDB) to integrate their Standard Emissions and Waste
Inventories (StEWI) with Factotum (for example, via migration to an updated open-source SQL
system and development of workflows for chemical curation) for delivery of release data to
internal or external stakeholders via existing ORD tools. The environmental occurrence data
curated into these databases (MMDB, Factotum, and StEWIDB) will inform the development of
much-needed predictive models of chemical releases into environmental compartments
assessment of TSCA-relevant chemicals.

3.	Toxicokinetic data for internal exposure (Concentration versus Time Database, CvTdb, and
high-throughput toxicokinetics, HTTK): Toxicokinetic data when combined with generic
toxicokinetic models provide a quantitative link between potential human exposures and
bioactive concentrations in in vitro screening. In addition, toxicokinetic data has the potential to
inform analog selection (e.g., selection of analogs with similar toxicokinetic properties) and

27


-------
BOSC Review Draft

refine chemical categories based on similar toxicokinetic properties among chemicals. Data
curated in CvTdb (in vivo tissue Concentration vs. Time database) enable quantitative evaluation
of confidence of HTTK predictions for many chemicals and provides data to refine generic HTTK
models (Sayre et al., 2020). Work planned in StRAP4 includes expansion of curated data for key
exposure routes (e.g., dermal and inhalation exposures) for TSCA-relevant chemicals that will be
useful in evaluating the generic HTTK models of dermal and inhalation exposures. Carefully
quantifying the chemical- and scenario-specific uncertainty in data and models allows decision
makers to consider the use of HTTK for IVIVE, as well as any other situation where extrapolation
using TK may be useful. This task is not possible without generating and continually expanding
the CvTdb resource, particularly to include routes of exposure and chemistries relevant to TSCA
for the NCCRP.

3. Develop and Refine (Q)SAR and Predictive Models for Physicochemical Properties,
Environmental Fate/Transport, Hazard, Exposure, and Toxicokinetics

OPPT has developed and applied a large suite of (Q)SAR and other predictive models, including
expert systems, to estimate physicochemical properties, exposure, environmental fate/transport,
carcinogenic hazard, and ecological hazard.21 OPPT and ORD are proposing to update and/or improve
existing OPPT (Q)SAR and predictive models and enable regular model updates. The data used to
develop and update (Q)SAR and predictive models will be derived from the curated public and TSCA
databases described above in Research Area 2. ORD and OPPT will evaluate all appropriate models,
including evaluation of the data used to build models and model performance against measured data, to
ultimately determine the best suite of models for use by OPPT for regulatory purposes.22

The OECD principles for validation of (Q)SAR (OECD, 2007) suggest that consideration of (Q)SARs for
regulatory purposes is facilitated by association of the (Q)SAR with information regarding the defined
endpoint predicted; a clear (and reproducible) algorithm; a defined domain of applicability; appropriate
measures of its goodness-of-fit, robustness, and predictivity; and, a mechanistic interpretation when
possible. ORD continues to pursue model development aligned with these principles, including use of
computational approaches such that updated model training and test data, critical to defining the

21	Predictive Models and Tools for Assessing Chemicals under TSCA

22	Using the OECD principles for validating QSAR models (see OECD 2007)

28


-------
BOSC Review Draft

applicability domain and predictivity, can be applied more rapidly and systematically. The ability to
rapidly extend the model training and test data to include more TSCA-relevant chemicals may increase
the applicability domain, thereby increasing the relevance of (Q)SAR predictions for new chemical
evaluations. In addition to extending model training and test data, evaluating applicability domain, and
updated reporting on model predictivity, ORD may also incorporate newer machine learning-based
approaches in model development. Whenever practicable, these new (Q)SARs will be contextualized
with existing models to build confidence and increased understanding across approaches. This will
include working with other stakeholders and peer reviewers to build confidence that the models related
to hazard or toxicokinetics meet the TSCA statutory requirement of Section 4(h)(l)(B)(i) to encourage
and facilitate "the use of scientifically valid test methods and strategies that reduce or replace the use of
vertebrate animals while providing information of equivalent or better scientific quality and relevance
that will support regulatory decisions...."

The goal of this effort is to update the models to reflect the best available science, increase
transparency, and establish a process for updating these models as science allows. This will enhance the
capabilities of OPPT to perform risk assessments for new chemicals. In addition, refining and developing
such tools will lead to their use by EPA, submitters, and other stakeholders in designing safer chemicals,
and will build confidence in their use for regulatory purposes. This research builds upon ORD efforts to
predict properties on the basis of structure for a variety of applications (Table 4).

Table 4. StRAP Outputs Relevant to (Q)SAR and Prediction

Properties predicted

Relevant StRAP4
Output

Relevant StRAP4 Output Title

Chemistry, hazard, and
properties

CSS.407.3

Develop new and improve existing structure
activity relationship models to support risk
assessment

Functional use and
exposure

CSS.402.3

Refine exposure models that enable high-
throughput exposure predictions for chemicals

Functional use and
exposure

CSS.408.4

Innovating science to support new chemicals
evaluation

Repeat dose point-of-
departure

CSS.408.4

A. Informatics platform for (Q)SAR development, implementation, and data management

Work in StRAP4 will extend ongoing ORD research on (Q)SAR modeling to inform chemical
evaluation, using a foundation of modeling best practices (e.g., the OECD (Q)SAR framework), updated

29


-------
BOSC Review Draft

methodologies, and input datasets. Where feasible, the predictive performance of resultant models will
be compared with current, peer-reviewed models being used by OPPT (e.g., EPI Suite, ECOSAR,
OncoLogic, OECD (Q)SAR Toolbox structure-based profilers)23 to ensure fit for purpose application and
build confidence in newer (Q)SARs. In addition, there is a need to establish cheminformatics approaches
for model management and versioning to enable real-time model predictions, data provenance, and
long-term sustainability to include ability to update, manage versions, and reproduce (Q)SAR values.

Such cheminformatics approaches also include development of automated workflows to transform raw
experimental data to modeling data sets and then optimize and streamline (Q)SAR model development.

Currently, WebTESTl.O is accessible through the CompTox Chemicals Dashboard for prediction24 of a
number of physicochemical and toxicological properties, such as oral rat acute 50% lethal dose. These
predictions are accessible by locating a structure in the CompTox Chemicals Dashboard or by drawing it
for real-time prediction. These models and their performance reports, including how the predicted
chemical fits within the applicability domain (in line with OECD (Q)SAR validation principles), are
available through the user interface. The cheminformatics approaches to be developed in StRAP4
represent ongoing and planned work to develop WebTEST2.0. This updated version of WebTEST will
include access to raw experimental data, modeling datasets, molecular descriptor values, and (Q)SAR
models.. The new WebTEST2.0 workflow can be used to develop models using Python-based machine
learning methods such as random forest and support vector machines, all within the WebTEST platform.

The impact of the work is that users will have access to real-time predictions from a large array of
(Q)SAR models, as well as other models run at defined intervals, in a single website. WebTEST2.0
predictions will be linked to an HTML report which indicates whether the chemical is within the
applicability domain of the models and provides prediction results for structurally similar chemicals from
the training and test sets. WebTEST 2.0 also provides extensive documentation on each (Q)SAR model.
For each property (and associated dataset), the tool will link to a spreadsheet which provides the
training and test set statistics, the training and test sets, and the test set prediction results for each
model. For each model, the tool will provide a PDF document in the (Q)SAR model reporting format that
outlines all the details of the model. Having a (Q)SAR model reporting format document is often a
requirement for using models for regulatory applications. Models for physicochemical properties (e.g.,
octanol water partition coefficient (logKow), vapor pressure, and Henry's law constant) are being

23	Predictive Models and Tools for Assessing Chemicals under TSCA

24	httpsi//comptox.epa.gov/dashboard/predictions

30


-------
BOSC Review Draft

developed using the WebTEST2.0 workflow. Revised toxicity models will be developed by expanding the
toxicity datasets for WebTESTl.O (e.g., acute aquatic toxicity). In addition, models will be developed for
additional toxicity endpoints (e.g., carcinogenicity, repeat dose toxicity, skin sensitization) to support
TSCA new chemical evaluations.

The architecture of WebTEST2.0 will also be extended to allow for incorporation of models which
are developed external to WebTEST2.0 using a different workflow. For example, a specialized fish
(Q)SAR model will be developed which is based on chemical classes obtained from the ClassyFire
webservice (Djoumbou Feunang et al., 2016). This model is essentially a more advanced version of the
fish toxicity model in EPI Suite. Models developed outside of the WebTEST platform will be implemented
via Docker containers or via API calls to external webservices. OPERA, EPI Suite, and WebTESTl.O models
will be incorporated into WebTEST2.0 via webservices. Additionally, bioactivity-based models for
estrogen receptor (Judson et al., 2017; Judson et al., 2015), androgen receptor (Judson et al., 2020;
Kleinstreuer et al., 2017), steroidogenesis (Haggard et al., 2018; Haggard et al., 2019), and potentially
other bioactivities based on in vitro NAM data, will be included in the WebTEST2.0 model registration
platform. Registration of all models, regardless of their development within or outside of the WebTEST
platform, will include meta-data on the input features used in the modeling, the model output, and
version information about that model; this constitutes an important goal for WebTEST2.0 and for rapid
integration of information from disparate sources for next generation risk assessment.

B. Exposure predictions

Review of new chemical submissions under TSCA includes both engineering assessment and
exposure assessment, including estimation of chemical releases to the workplace and environment
based on the chemical "conditions of use," as well as estimation of resulting occupational, general
population (ambient), and consumer exposures. In many cases, these estimates also include site-specific
exposures. While the process and exposure models used in these evaluations are well-defined and peer-
reviewed, there are several areas where ORD research efforts could improve or expand current
evaluation workflows (Wambaugh et al., 2019). The research efforts include systematic approaches for
identifying potential uses for submitted substances, estimating exposure via poorly
characterized release scenarios (e.g., down-the-drain consumer or industrial releases), characterizing
variability and uncertainty (which could inform identification of highly exposed populations), and
addressing data-poor chemicals (e.g., those for which no default information is available for
parameterizing likely exposure scenarios). Methods for characterizing potential conditions of use may

31


-------
BOSC Review Draft

utilize expanded functional use databases (to better cover the use space for known chemical analogs)
and refined (Quantitative) Structure Use Relationship ((Q)SUR) models that can consider a chemical
functional role within a particular sector (e.g., consumer versus industrial use) or specific industry or
product category. In addition, ORD research efforts will improve the flexibility of currently available
models to better characterize a wider range of exposure scenarios, including the consideration of spatial
scale and environmental justice concerns. ORD is planning research to deliver new computational
models, workflows, and datasets to support problem formulation (identification of exposure scenarios
relevant to a particular chemical) and/or quantitative modeling related to exposure assessment in the
PMN process. Where feasible, these newer research tools will be compared to the current approaches
and models used by OPPT.

Though not currently used in assessment of new chemical submissions under TSCA, ORD is engaged
in ongoing development of refined consensus exposure models that can use chemical structure
representations to predict: 1) human daily chemical intake rates; 2) air concentrations in occupational
settings; and 3) surface water concentrations. These models are examples of EPA's systematic empirical
evaluation of models (SEEM) meta-model approach (Ring et al., 2019) and integrate available predictors
from multiple exposure pathway models with available monitoring data to produce predictions for novel
chemical structures. These models can provide screening-level exposure estimates (with associated
uncertainty) for data-poor chemicals. When completed, SEEM exposure prediction models will be
integrated with the WebTEST2.0 cheminformatics platform as models developed outside of the
WebTEST platform.

C. Toxicokinetic predictions

ORD has developed a library of empirically measured HTTK parameters (that is, in vitro toxicokinetic
measurements) for >1000 chemicals. However, with tens of thousands of chemicals that may be of
interest, ORD has been engaged in development of a series of quantitative structure-property
relationship models for predicting key toxicokinetic parameters, including fraction of chemical unbound
to protein in plasma, and hepatic metabolic clearance (Dawson et al., 2021; Pradeep et al., 2020; Sipes
et al., 2017). Work is underway to compare different in silico methods for toxicokinetic parameter
prediction, and these approaches will be applied to TSCA-relevant chemicals to make IVIVE possible
even without experimental data for toxicokinetic parameters to inform generic toxicokinetic models.
Further, predicted toxicokinetic parameters inform models of bioavailability and dermal absorption and
may refine chemical categories and analog selection (i.e., considering neighboring chemicals with similar
toxicokinetic properties). The toxicokinetic parameter predictions, and the subsequent generic HTTK

32


-------
BOSC Review Draft

model predictions, will be integrated with the WebTEST2.0 cheminformatics platform as models
developed outside of the WebTEST platform.

4. Explore Ways to Integrate and Apply In Vitro NAMs in New Chemical Assessments

Section 4(h) of TSCA promotes reducing testing on vertebrate animals and sets forth some

requirements for NAMs. As recognized in OPPT's Strategic Plan,25 leveraging in vitro NAMs to generate
mechanistic, hazard, and toxicokinetic data may further inform data gap filling approaches for new
chemicals. As required under TSCA 4(h), OPPT maintains a list of NAMs that are scientifically reliable,
relevant, and capable of providing information of equivalent or better scientific reliability and quality to
that which would be obtained from vertebrate testing.26

EPA and the broader scientific community have invested heavily in the development of in vitro
NAMs. As part of the NCCRP, OPPT is proposing to take advantage of previous and ongoing research in
ORD and by other partners that have identified important biological targets representing potential
hazards, improved estimates of dose extrapolation from in vitro systems, incorporated routes that are
key to highly exposed populations (e.g., inhalation and dermal exposure), and continued to develop
resource effective technologies that broadly characterize biological activities across pathways,
processes, and different cell types. The in vitro NAMs applied within the NCCRP will be evaluated for
reliability and relevance for new chemical evaluation. Fit-for-purpose application of NAMs will rely, to
the extent possible, on the concepts of (1) adverse outcome pathways (AOPs) and the key events
leading to toxicity; and (2) Integrated Approaches to Testing and Assessment (IATA) for weight of
evidence evaluation and the use of Defined Approaches.27 Although some informative NAMs may not be
associated with an IATA or Defined Approach, and some health outcomes do not have established AOPs,
this would not prevent OPPT from applying these methods if they represent the best available science.

The proposed effort is intended to develop a suite of accepted, fit-for-purpose NAMs that could
be used by external stakeholders for testing and data submissions under TSCA as well as informing and
expanding new chemical categories. In this Research Area, ORD will collect in vitro NAM data to
demonstrate how NAMs for bioactivity and toxicokinetics can be used in a NAM-informed assessment of
data-poor chemicals. The NCCRP presents the opportunity for ORD to make a leap in progress on

25	See published plans by OPPT under TSCA (2018) and by EPA for the Agency (2021).

26	See TSCA Section 4(h) NAM list

27	AOP; see G Patlewicz et al. (2015). Proposing a scientific confidence framework to help support the application
of adverse outcome pathways for regulatory purposes. Regul. Toxicol. Pharmacol. 71(3):463-77. doi:
10.1016/j.yrtph.2015.02.01. IATA - see IATA and Defined Approaches (OECD 2017).

33


-------
BOSC Review Draft

prospective application of a screening strategy for hazard (Thomas et al. 2019) (see Figure 3 for outline
of screening work planned in StRAP4). In a first step, ORD will focus on development of a dataset for
200-300 chemicals, including some reference chemicals as well as TSCA-relevant chemicals from the
nonconfidential inventory, to increase scientific confidence in application of this suite of bioactivity
NAMs for informing chemical safety. These data will be needed to evaluate performance of these NAMs
for further application and may also inform evolving frameworks for using multiple data streams to
inform bioactivity-based dose-response assessment and hazard identification. Pending additional
infusion of resources, bioactivity screening could be extended to additional chemicals, which is a
necessary component of using bioactivity for analog selection or informing putative chemical categories
that cover a large TSCA-relevant chemical universe.

A cheminformatics step will identify candidate chemicals for screening from the TSCA active
inventory, including examination of: chemical structural and physicochemical diversity to promote
coverage of putative chemical categories; amenability to aqueous based-screening or potential volatility
or aerosolize-ability; ability to procure the chemical in sufficient quantities for screening; and, chemicals
of interest with respect to current gaps in bioactivity and/or (Q)SAR model data sets. Following chemical
selection, a set of 200-300 chemicals amenable to aqueous screening will be screened in both broad and
targeted biological screening technologies for human health relevant endpoints, and a subset of these
chemicals will be tested in broad screens with ecological relevance. "Broad" screening refers to
technologies such as high-throughput phenotypic profiling (HTPP) and high-throughput transcriptomics
(HTTr) that characterize the biological activity of chemicals using highly-multiplexed measurements of
many different cellular features or transcripts, respectively, thereby capturing chemical effects that may
result from specific interactions with molecular initiating event (MIE) targets as indicated by fingerprints
or signatures indicative of those MIEs, as well as effects that may result from generalized cellular
responses to stress or activity at multiple MIEs (Harrill et al., 2021; Nyffeler et al., 2020). These broad
screening modalities inform both estimates of in vitro bioactivity-based dose-response assessment and
identify whether a chemical may act at specific MIEs or non-specifically at many targets (Thomas et al.,
2019). Targeted screening complements broad screening by providing information about MIEs, key
events, or other processes related to hazards of interest. Broad and targeted screens are combined in
ORD work to support the NCCRP for human and ecological health as well as for acute portal of entry
effects from potential inhalation exposures. A smaller subset of chemicals with inhalation exposure
potential (8-10 chemicals per year) will be screened in NAMs for inhalation exposure that include both
broad and targeted measures (Figure 3 for outline).

34


-------
BOSC Review Draft

Figure 3. Overview of an initial in vitro NAM screening strategy.

Cheminformatic approaches will be used to categorize the structural and physicochemical diversity of
the TSCA active inventory and identify potential chemical screening candidates; these candidates will be
further refined to those that are procurable and amenable to aqueous-based screening in cell-based and
cell-free assays (including physicochemical property evaluation and non-volatility) or cell-based models
of inhalation with an air:liquid interface. Candidate chemicals would also fill gaps in available in vitro or
in silico screening information (e.g., filling gaps in the applicability domain for ECOSAR). Broad and
targeted screening covering human and ecological health and inhalation exposure will be applied. Broad
screens are represented by HTPP also known as Cell painting for human health, tiered application of
modified ecotoxicology studies and HTTr for ecotoxicology (EcoHTTr), and HTTr applied to multiple cell
models of the human respiratory tract in an air:liquid interface system. Targeted screens are
represented by a safety pharmacology panel (Safety Pharm), a DevTox Germ Layer Reporter assay,
assays for genotoxicity, and assay data to inform high-throughput toxicokinetics (HTTK). Phenotypic
measures in models of human respiratory tract also represented targeted screens conducted in parallel
to HTTr in the inhalation exposure system. Inverted triangles represent "funnels" to indicate relative
numbers of chemicals to be screened in the initial strategy out of the 200-300 chemicals selected.

35


-------
BOSC Review Draft

Table 5. StRAP Outputs Relevant to In Vitro NAM Data Generation for NCCRP

Bioactivity screening planned

Relevant

StRAP4

Output

Relevant StRAP4 Output
Title

Analytical quality control

CSS.408.4

Innovating science to

Developmental toxicity screening



support new chemicals
evaluation

Acute portal of entry effects of inhaled exposures





Toxicokinetic NAMs (hepatic clearance and fraction
unbound)





Broad screening (high-throughput phenotypic
profiling; safety pharmacology)

CSS.408.4,
CSS.401.1

Additional reliance on:
Advance a tieredhigh-
throughput toxicity testing
strategy

Ecological hazard screening

CSS.408.4,
CSS.406.5

Additional reliance on:
Improve ecological
methods and models for
predicting exposure,
accumulation, and effects
of P FAS

A. Analytical quality control of chemicals

For aqueous-based screening assays for human and ecological health, ORD will procure chemical
samples, solubilize in dimethyl sulfoxide, and characterize the identity and purity of the samples using
liquid chromatography or gas chromatography, as appropriate, with tandem mass spectrometry. This
analytical quality control step will ensure that the learnings from the 200-300 chemical set screened in
the cell-based and cell-free assays will be more interpretable with respect to the nominal concentration
and identity of chemicals screened. For volatilized samples used in the inhalation exposure model, air
samples are collected just before reaching the exposure chamber by syringe and directly transferred to a
gas chromatograph for characterization of the chemical and its concentration.

B. Screening for human health

Broad-based biological screens will be employed to provide insight into as many putative chemical-
by-biological target interactions as possible, as suggested by the use of Tier 1 NAMs in the CompTox
BluePrint (Thomas et al., 2019). These broad profiling assay data could potentially provide a basis for

36


-------
BOSC Review Draft

derivation of a bioactivity-based point of departure (POD) (Baltazar et al., 2020; Nyffeler et al., 2020;

Paul Friedman et al., 2020) as well as coverage of specific MIEs, indicated by the fingerprint or pattern of
behavior in these assays, that may be of interest for ascertaining the need for additional hazard
information. One such Tier 1 NAM, high-throughput phenotypic profiling (HTPP), also known as Cell
Painting, measures more than 1000 cellular morphological features using high content imaging to inform
a quantitative POD, putative molecular targets or molecular initiating events, and bioactivity fingerprints
that could be used to evaluate similarity in biological effects measured by this assay (Nyffeler et al.,
2021; Nyffeler et al., 2022; Nyffeler et al., 2020; Willis et al., 2020).

Targeted NAMs for hazard will also be important to informing human safety assessment gaps left by
broad profiling NAMs. Safety pharmacology has been employed to detect off-target interactions in drug
safety (Bowes et al., 2012; Papoian et al., 2017; Smit et al., 2021) and cosmetic products under a next
generation risk assessment framework (Baltazar et al., 2020). There appears to be consensus that a
broad and diverse panel of pharmacological targets can be useful information for drug safety
assessment (letswaart et al., 2020; Valentin et al., 2018). A similar panel of receptor, ion channel,
transporter, and enzyme assays has been employed previously within the larger set of assays included in
the US EPAToxCast program (Sipes et al., 2013). HTPP in multiple cell lines, each expressing a different
suite of potential molecular targets, will be combined with a large panel of nuclear receptor, G-protein-
coupled receptor, transporter, ion channel, and enzymatic target assays to ensure coverage of MIEs
known to be of interest for toxicology.

Another important hazard gap is genotoxicity; commercially available Ames and in vitro
micronucleus assays will be employed to evaluate genotoxicity potential of the 200-300 chemical set.
Developmental toxicity potential is rarely evaluated in data submitted or associated with data-poor new
chemical submissions to OPPT. To evaluate developmental toxicity potential, ORD will apply an assay
adaptation of the human pluripotent stem cell test (Kameoka et al., 2014), which identifies potential
developmental toxicants using a biomarker of early embryonic development. The assay adaptation
employs the human RUES2-GLR stem cell line that has been engineered with a fluorescent reporter for
the SOX17 biomarker to monitor differentiation of the endoderm germ layer. Recently, the assay
protocol has been optimized to a 384-well plate format with shortened exposure duration to enable
more rapid high-throughput screening (Gamble et al., 2022). The assay demonstrates the ability to
distinguish assay negatives such as acetaminophen from known developmental toxicants such as
thalidomide. Applying this cell-based developmental toxicity germ layer reporter assay (DevTox GLR-

37


-------
BOSC Review Draft

Endo assay) to a 200-300 chemical set would provide the data necessary to demonstrate utility not only
for the NCCRP and needs of TSCA but also for broader acceptance and validation. The in vitro
developmental toxicity data, combined with (Q)SAR, read-across, and chemical categories, may provide
information that is typically not available from in vivo studies for new chemical submissions under TSCA.

In vitro toxicokinetic assays for metabolic intrinsic clearance and plasma protein binding will also be
conducted for the 200-300 chemical data set. As mentioned above in Research Area 3, empirical (and in
silico) toxicokinetic parameters can be used to inform generic HTTK models for IVIVE, predictions of
bioavailability and dermal absorption, and potentially, as inputs into characterization of chemical
categories or analogs that might share similar toxicokinetic properties. Combining in vitro assay data
from these and similar assay platforms with in vitro distribution and toxicokinetic modeling will increase
the utility of these data by providing improved estimates of doses that correspond to in vitro bioactivity
in exposure units that can be compared to in vivo dose estimates or exposure estimates (Honda et al.,
2019; Klaren et al., 2019; Ring et al., 2021).

C. Screening for ecological health

Ecological hazard for new chemicals has focused on potential toxicity to three representative groups
of aquatic organisms: fish, invertebrates, and plants/algae. Data for these three taxonomically diverse
groups is often lacking for new chemical submissions, and as such, new chemical assessments have
relied heavily on read-across to structural analogs with available toxicity estimates or (Q)SARs such as
ECOSAR that rely largely on physicochemical properties of the new chemical such as its relative
lipophilicity. ORD proposes to work with OPPT to identify a set of up to 60 chemicals that represent five
chemical structural domains of interest for which ecological toxicity data and/or understanding of the
applicability of currently (Q)SAR models is limited. Selected chemicals will be screened using higher-
throughput adaptations of guideline in vivo ecotoxicity assays to estimate no and lowest observable
effect concentrations for fish and invertebrates, 50% lethal concentration for fish and invertebrates, as
well as a 50% effect concentration for algae for comparison with the values predicted by ECOSAR. For
cases where current (Q)SAR approaches do not appear predictive, a smaller subset of the chemicals
would then be selected for additional screening using ecological high throughput transcriptomics (Eco-
HTTr) assays with fish, invertebrates, and/or plants as relevant. Data would be used to both derive a
transcriptomics-based POD and support mechanistic inference for the tested chemicals.

38


-------
BOSC Review Draft

D.	Screening for inhalation exposures

For chemicals with potential inhalation exposures, specialized cellular architectures to more closely
recapitulate human biology are needed, but the throughput on these systems currently prevents
screening large numbers of chemicals. An engineering and exposure assessment for a data-poor
substance under TSCA may suggest the potential for inhalation exposure, but frequently limited to no
inhalation data may be available for assessing hazard of these exposures for the target chemical. Read-
across, category-based approaches, and adaptation of the threshold of toxicological concern for
inhalation toxicity can be leveraged to understand hazard based on inhalation, but all of these
approaches rely on existing information on well-characterized chemicals. ORD plans to continue
developing and applying in vitro methods for evaluating acute portal of entry effects of inhaled chemical
exposures as part of the NCCRP to demonstrate the utility of this approach and its transferability (Zavala
et al., 2017; Zavala et al., 2018). The prospective view is that ORD would screen 8-10 chemicals per year
in multiple cell models of the human respiratory tract, pending sufficient resources, and that these data
would inform read-across and other in silico approaches that would extend the impact of limited
empirical screening. These studies combine ORD-developed in vitro exposure technology (the Cell
Culture Exposure System), organotypic in vitro cell-based models of the human respiratory tract, and an
assay battery that evaluates phenotypically relevant endpoints (i.e., targeted screens) such as
cytotoxicity, cellular metabolism, epithelial barrier function, mucin production, ciliary beat frequency,
and cytokine production. In addition, high-throughput transcriptomic (HTTr) evaluation in models of
human respiratory tract represents a broad screening technology to identify biological pathway activity
and a benchmark concentration, increasing the richness of the information and sensitivity for acute
portal of entry effects (Speen et al., 2022). Importantly, generation of this dataset for additional TSCA-
relevant chemicals, including PFASthat have not been tested in an in vitro air-liquid interface system
previously, with potential inhaled exposures will inform guidance to support deployment of this in vitro
NAM for chemical evaluations by external partners and stakeholders.

E.	Additional bioactivity data

Additional bioactivity data may be generated in other endeavors within the StRAP, including HTTr
screening work in other ORD CSS research under High-throughput Toxicity Testing (CSS 401.1 and CSS
401.2, see Figure 2), for the 200-300 chemical case study, depending on resources allocated. When
additional bioactivity data are available for these 200-300 chemicals from other assays, work to

39


-------
BOSC Review Draft

demonstrate the translation of these data to meaningful biological or quantitative POD estimates may
be informative for future NCCRP work products.

5. Develop a TSCA New Chemicals Decision Support Tool to Modernize the Process

Within OPPT, searching, collating, and integrating data on new chemicals is inefficient and
hinders the timeliness of decision-making. The international regulatory community has been moving
towards using IUCLID to capture, store, maintain, and exchange data on intrinsic and hazard properties
of chemical substances. Data in IUCLID require standardized reporting templates; for many data types,
these reporting templates are consistent with internationally accepted test guidelines. ORD is proposing
to use IUCLID to capture, store, and maintain publicly available data on intrinsic and hazard properties
and exposure-related data. These efforts will promote data interoperability between OPPT, ORD, and
other stakeholders.

Available digitized data for TSCA chemicals is important for delivery of a decision support tool
that integrates all the data streams (e.g., chemistry, fate, exposure, hazard) for risk assessment and
transparently documents the decisions and assumptions made by expert users based on available
information. This will facilitate NCD tracking decisions over time and evaluating consistency within and
across chemistries. OPPT and ORD propose to collaborate on identifying the appropriate content and
workflow for such a decision support tool. For example, the proposed decision support tool may allow
expert chemists to examine the types of data available for analogs for a target. Information on chemical
categories and/or analogs, and estimates of physicochemical properties, environmental fate, hazard,
and toxicokinetics generated from predictive and in vitro models, will be included in the decision
support tool, thereby limiting time spent on manual searching, compiling, and contextualizing available
information and enabling more rapid and reproducible decisions over time.

The work to support Research Area 5 has three main components: (1) OPPT and ORD
collaboratively working to increase the amount of computationally accessible CBI data previously
submitted to OPPT for use within a CBI-protected environment; (2) ORD bringing computationally
accessible and public data, some of which is already in ORD databases, into lUCLID-compatible formats
that enable collation of these data with other data in IUCLID format; and, (3) development of a proof-of-
concept decision support tool that consolidates as much traditional and NAM data as possible into a
single workflow to inform assessments and documentation of selections and assumptions, such as data
for decisions. This work, spanning multiple StRAP Outputs (Table 6), is intended to begin addressing the

40


-------
BOSC Review Draft

challenge OPPT NCD faces in searching for, collating, and integrating data for a new chemical
assessment.

Table 6. StRAP Outputs Relevant to IUCLID and a NCCRP decision support tool

Relevant StRAP4
Output

Relevant StRAP4 Output Title

CSS.408.1

Integrating data systems to enable knowledge delivery

CSS.408.3

Cross-disciplinary integration and applied case studies to support chemical
safety decision making

CSS.408.4

Innovating science to support new chemicals evaluation

A.	Implementing the International Uniform Chemical Information Database (IUCLID) in ORD

Regulatory authorities including the European Chemicals Agency, the European Food Safety Agency,
and Health Canada employ IUCLID database formatting, which uses standardized reporting templates to
manage chemical data (i.e., Organisation for Economic Co-operation and Development [OECD]
Harmonized Templates, known as OHTs) (OECD, 2021). IUCLID is an international effort with dedicated
resources to manage, update, and develop tools around IUCLID, such as the Data Uploader tool to
convert data to IUCLID format. ORD has extensive ongoing data curation efforts in customized database
schemas to capture research-oriented levels of detail. In support of NCCRP as well as ongoing internal
and external collaboration, wherever possible based on existing OHTs, these databases will be mapped
to OHTs for conversion to IUCLID format, with initial priorities on physicochemical properties and human
and ecological health data. A stable computing environment for IUCLID will be established in ORD.
Development of automated processing of data in the ORD research environment to IUCLID will enhance
ORD's ability to use these public IUCLID data in a decision support tool for new chemicals as well as
other modeling applications with internal and external stakeholders. Further, ORD would be able to
transfer IUCLID formatted public data to operate within the TSCA CBI instance of IUCLID, such that a
combined IUCLID dataset could supply one of the data streams for a proof-of-concept decision support
tool.

B.	Collaboration between ORD and OPPT on IUCLID data

Though OPPT is able to receive data in IUCLID, historical data and the majority of incoming data are
not reported in OHT formats and may exist as documents or digitized data stored in disparate locations.
Modernizing the new chemical assessment process involves being able to rapidly exchange data, which

41


-------
BOSC Review Draft

is supported by migrating as much of these data as possible via OHT formats to IUCLID. In work
complementary to the NCCRP, OPPT is working to receive new chemical data submissions in IUCLID
format or convert submitted study information to OHT-compliant formats and relay to IUCLID. Ongoing
efforts within OPPT aim to reformat and migrate the data submitted under TSCA from current databases
to an instance of IUCLID in OPPT's protected CBI environment. ORD is supporting this effort via pilot
work to digitize some OPPT data, including publicly available TSCA Section 8(e) reports, as well as new
chemical assessment reports and other summary hazard documents within OPPT's protected CBI
environment. Overall, this collaboration to support digitization, integration, and ultimate conversion to
lUCLID-compatible formats will support collation of these data for utilization by applications, such as a
decision support tool.

C. Developing proof-of-concept decision support tool for new chemicals

Next generation risk assessment workflows that bring together chemistry, hazard, toxicokinetics,
and exposure demonstrate early efforts in the open literature to create reproducible analyses and
consolidate disparate data for regulatory toxicology (Baltazar et al., 2020; Beal et al., 2022; Dent et al.,
2021; Ouedraogo et al., 2022; Paul Friedman et al., 2020; Rajagopal et al., 2022; van Tongeren et al.,
2021). In ORD, CCTE is building an ecosystem of decision support tools to meet stakeholder needs and
advance toward next generation risk assessments that utilize as much computationally accessible
information as possible. To enable OPPT to rapidly review relevant information from both traditional
and NAM sources and make reproducible and documented decisions using many types of information, a
proof-of-concept decision support tool will be developed as part of this ecosystem. This decision
support tool will integrate data domains such as chemistry, hazard (including data from IUCLID),
bioactivity (including IVIVE of dose), environmental fate, functional use, and exposure. Development of
this tool requires expertise and teamwork from regulatory experts in OPPT NCD, ORD technical experts
in the data domains, and experts in software development, data engineering, systems administration,
among other information technology domains. The intention within ORD is to develop this tool in a
modular, rapid, and innovative way with feedback from OPPT. Continual assessment of how to best
mature the application within CCTE's ecosystem of decision support tools and data architecture will also
be needed. A draft overview of a possible proof-of-concept decision support tool is illustrated in Figure
4.

42


-------
BOSC Review Draft

Figure 4. Draft overview of proof of concept for NCCRP decision support tool.

Expert users would (1) define a target structure; (2) explore chemical similarity among possible analogs
using structural descriptors, substructure, physicochemical and fate properties, metabolite profile,
and/or bioactivity response profile and select analogs based on available exposure and/or hazard
information for these similar analogs; (3) explore the exposure data landscape for selected analog(s),
including functional use and exposure/fate data and predictions; (4) explore the hazard data landscape
for selected analog(s), including (Q)SAR and structure alerts, in vitro bioactivity, and in vivo hazard data,
with the aim of identifying key toxicity types and/or a POD; (5) explore the hazard landscape for
ecological health, including (Q)SARs and in vivo data relevant to a POD; (6) explore integrated toxicity
views for select effects such as developmental and reproductive toxicity (DART) or specific organ toxicity
to build a weight of evidence or fill data gaps for these toxicity types; and (7) generate reports that
include selected data used, narrative justifications for decisions, and automated summaries of the data.

Proof of Concept Decision Support Tool Workflow

^ Explore chemical similarity to select analogs

Structural descriptors

Physicochemical and
fate properties

Toxicokinetic and
metabolite profile
Bioactivity response
profile

Exposure info
for analog(s)

Hazard info
for analog(s)

Exposure Data
Landscape

(Q)SUR and Exposure
Functional use and
(Q)SUR

Exposure and Fate

[^Hazard Data Landscape

PODs and Key Effects
(Q)SAR and
structure alerts
In vitro
bioactivity, PODs

In vivo hazard

5 Ecological Data
Landscape

PODs and Key Effects
(Q)SAR and
structure alerts

In vivo hazard

1 Integrated Toxicity
Views

Explore select effects, e.g.
"DART" or "Liver"
Select (Q)SAR and
structure alerts
Select in vitro data

for toxicity type
Select in vivo data
for toxicity type

|7 Report Generation

Record selections, tabular
outputs, summaries

In support of these goals for the NCCRP, ORD plans to extend ongoing work on the Cheminformatics
Analysis Modules, a proof-of-concept decision support tool that currently integrates data streams
including curated in vivo hazard data, structural alerts, predicted and experimental physicochemical and
environmental fate and transport properties, as well as (Q)SAR-predicted toxicity endpoints, and
ToxPrint chemotypes. Important needs in the NCCRP are: (1) to connect experts in NCD with analog-
related data as chemical submissions are mostly data poor; (2) organize/output data for easy
interpretation and incorporation into new chemical evaluations; (3) add functionality to connect
structures and/or their analogs to functional uses, and environmental fate and exposure data; (4) add
functionality to evaluate bioactivity and integrate targeted bioactivity (from Research Area 4) with

43


-------
BOSC Review Draft

structural alerts, (Q)SARs or predictive models for hazard, and/or any available hazard data for specific
toxicity types (e.g., developmental or reproductive toxicity and carcinogenicity), and (5) add
functionality to connect analog information to any available (Q)SAR, structure alerts, or relevant in vivo
hazard information relevant to ecological health. Each proof-of-concept module will need to export
and/or capture selection of data for an assessment.

A draft overview of a possible proof-of-concept decision support tool, based on extension of the
Cheminformatics Analysis Modules, is illustrated in Figure 4. Additional features for this proof-of-
concept decision support tool could include a structure searching and chemical profiling module, which
would provide various alerting rules (e.g., membership in lists such as IARC collections, Ashby
carcinogenicity alerts, Threshold of Toxicological Concern (TTC) alerts, and other user-definable
approaches). Additionally, physicochemical and metabolite predictions may be generated. Read-across
based on similarity metrics informed by structure, properties, and/or available data may be included as
a module. Another module may include data and/or predictions for exposure, fate, and functional use; a
set of modules could review the hazard data landscape for human and ecological health, including
(Q)SAR predictions and category or analog approaches to identifying relevant in vivo and in vitro hazard
data; and another set of modules could include integrated toxicity views, or views of structure alerts,
(Q)SARs, applicable in vitro data and in vivo data for specific toxicity types, such as developmental and
reproductive toxicity. Importantly, this workflow would provide reporting capabilities within each
module to support OPPT in developing their assessments. Users will be able to save selections of
information considered important for the new chemical assessment along with any narrative
justifications. Developing this tool will allow ORD and OPPT to gain experience working together as they
iteratively refine design requirements. As a long-term goal, ORD and OPPT will be creating a software
tool with key functionality that can be populated with public and/or CBI information in a CBI-protected
environment to improve the overall workflow and decision-making process for NCD chemical
assessments.

Conclusion

The proposed research relevant to the NCCRP in the FY23-26 StRAP is extensive, connecting with
many of the goals in ORD to support next generation risk assessment through the development and
implementation of NAMs and decision support tools. Collaboration with OPPT will ensure that this
research leads to fit-for-purpose translation and implementation, and that the needs of regulatory
decision-making influence the research in ORD. Though the ultimate success of the proposed research in

44


-------
BOSC Review Draft

ORD is resource dependent, the vision laid out in this research program will not only create proof-of-
concept next generation risk assessment tools for OPPT, but will also demonstrate progress toward
accomplishing key goals in the EPA NAM Work Plan and the CompTox BluePrint.

References

Baltazar, M. T., et al., 2020. A Next-Generation Risk Assessment Case Study for Coumarin in Cosmetic
Products. Toxicol Sci. 176, 236-252.

Beal, M. A., et al., 2022. Implementing in vitro bioactivity data to modernize priority setting of chemical
inventories. ALTEX. 39, 123-139.

Bowes, J., et al., 2012. Reducing safety-related drug attrition: the use of in vitro pharmacological
profiling. Nat Rev Drug Discov. 11, 909-22.

Dawson, D. E., et al., 2021. Designing QSARs for Parameters of High-Throughput Toxicokinetic Models
Using Open-Source Descriptors. Environ Sci Technol. 55, 6505-6517.

Dent, M. P., et al., 2021. Paving the way for application of next generation risk assessment to safety
decision-making for cosmetic ingredients. Regul Toxicol Pharmacol. 125, 105026.

Dionisio, K. L., et al., 2018. The Chemical and Products Database, a resource for exposure-relevant data
on chemicals in consumer products. Sci Data. 5, 180125.

Djoumbou Feunang, Y., et al., 2016. ClassyFire: automated chemical classification with a comprehensive,
computable taxonomy. J Cheminform. 8, 61.

Gamble, J. T., et al., 2022. The DevTox Germ Layer Reporter Platform: An Assay Adaptation of the
Human Pluripotent Stem Cell Test. Toxics. 10.

Grulke, C. M., et al., 2019. EPA's DSSTox database: History of development of a curated chemistry
resource supporting computational toxicology research. Comput Toxicol. 12.

Haggard, D. E., et al., 2018. High-Throughput H295R Steroidogenesis Assay: Utility as an Alternative and
a Statistical Approach to Characterize Effects on Steroidogenesis. Toxicol Sci. 162, 509-534.

Haggard, D. E., et al., 2019. Development of a prioritization method for chemical-mediated effects on
steroidogenesis using an integrated statistical analysis of high-throughput H295R data. Regul
Toxicol Pharmacol. 109, 104510.

Harrill, J. A., et al., 2021. High-Throughput Transcriptomics Platform for Screening Environmental
Chemicals. Toxicol Sci. 181, 68-89.

Helman, G., et al., 2019a. Quantitative prediction of repeat dose toxicity values using GenRA. Regul
Toxicol Pharmacol. 109, 104480.

Helman, G., et al., 2018. Extending the Generalised Read-Across approach (GenRA): A systematic

analysis of the impact of physicochemical property information on read-across performance.
Comput Toxicol. 8, 34-50.

Helman, G., etal., 2019b. Generalized Read-Across (GenRA): A workflow implemented into the EPA
CompTox Chemicals Dashboard. ALTEX. 36, 462-465.

Honda, G. S., et al., 2019. Using the concordance of in vitro and in vivo data to evaluate extrapolation
assumptions. PLoS One. 14, e0217564.

letswaart, R., et al., 2020. Machine learning guided association of adverse drug reactions with in vitro
target-based pharmacology. EBioMedicine. 57, 102837.

Isaacs, K. K., et al., 2022. A harmonized chemical monitoring database for support of exposure
assessments. Sci Data. 9, 314.

45


-------
BOSC Review Draft

Judson, R., et al., 2020. Selecting a minimal set of androgen receptor assays for screening chemicals.
Regul Toxicol Pharmacol. 117, 104764.

Judson, R. S., et al., 2017. On selecting a minimal set of in vitro assays to reliably determine estrogen
agonist activity. Regul Toxicol Pharmacol. 91, 39-49.

Judson, R. S., et al., 2015. Integrated Model of Chemical Perturbations of a Biological Pathway Using 18
In Vitro High-Throughput Screening Assays for the Estrogen Receptor. Toxicol Sci. 148, 137-54.

Kameoka, S., et al., 2014. A high-throughput screen for teratogens using human pluripotent stem cells.
Toxicol Sci. 137, 76-90.

Klaren, W. D., et al., 2019. Identifying Attributes That Influence In Vitro-to-ln Vivo Concordance by
Comparing In Vitro Tox21 Bioactivity Versus In Vivo DrugMatrix Transcriptomic Responses
Across 130 Chemicals. Toxicol Sci. 167, 157-171.

Kleinstreuer, N. C., et al., 2017. Development and Validation of a Computational Model for Androgen
Receptor Activity. Chem Res Toxicol. 30, 946-964.

Mansouri, K., et al., 2018. OPERA models for predicting physicochemical properties and environmental
fate endpoints. J Cheminform. 10,10.

Mansouri, K., et al., 2016. An automated curation procedure for addressing chemical errors and

inconsistencies in public datasets used in QSAR modelling. SAR QSAR Environ Res. 27, 939-965.

Nyffeler, J., et al., 2021. Comparison of Approaches for Determining Bioactivity Hits from High-
Dimensional Profiling Data. SLAS Discov. 26, 292-308.

Nyffeler, J., et al., 2022. Combining phenotypic profiling and targeted RNA-Seq reveals linkages between
transcriptional perturbations and chemical effects on cell morphology: Retinoic acid as an
example. Toxicol Appl Pharmacol. 444, 116032.

Nyffeler, J., et al., 2020. Bioactivity screening of environmental chemicals using imaging-based high-
throughput phenotypic profiling. Toxicol Appl Pharmacol. 389, 114876.

OECD, Guidance Document on the Validation of (Quantitative)Structure-Activity Relationships (QSAR)
Models. ENV/JM/MONO(2007)2. Environment Directorate, Joint Meeting of the Chemicals
Committee and the Working Party on Chemicals, Pesticides, and Biotechnology,
https://www.oecd.org/officialdocuments/publicdisplavdocumentpdf/?doclanguage=en&cote=e
nv/im/monof200712. 2007.

OECD, Guidance on the Grouping of Chemicals, Second Edition. Vol. No. 194. Environment Directorate,
Joint Meeting of the Chemicals Committee and the Working Party on Chemicals, Pesticides, and
Biotechnology, ENV/JM/MONO(2014)4, 2014.

OECD, Customisation Opportunities of IUCLID for the Management of Chemical Data - 2nd Edition, OECD
Series on Testing and Assessment, https://doi.org/10.1787/	76. OECD Publishing, Paris,

2021.

Olker, J. H., et al., 2022. The ECOTOXicology Knowledgebase: A Curated Database of Ecologically

Relevant Toxicity Tests to Support Environmental Research and Risk Assessment. Environ Toxicol
Chem. 41, 1520-1539.

Ouedraogo, G., et al., 2022. Read-across and new approach methodologies applied in a 10-step
framework for cosmetics safety assessment - A case study with parabens. Regul Toxicol
Pharmacol. 132, 105161.

Papoian, T., et al., 2017. Regulatory Forum Review*: Utility of in Vitro Secondary Pharmacology Data to
Assess Risk of Drug-induced Valvular Heart Disease in Humans: Regulatory Considerations.
Toxicol Pathol. 45, 381-388.

Paul Friedman, K., et al., 2020. Utility of In Vitro Bioactivity as a Lower Bound Estimate of In Vivo Adverse
Effect Levels and in Risk-Based Prioritization. Toxicol Sci. 173, 202-225.

Phillips, K. A., et al., 2017. High-throughput screening of chemicals as functional substitutes using
structure-based classification models. Green Chem. 19, 1063-1074.

46


-------
BOSC Review Draft

Pradeep, P., et al., 2020. Using Chemical Structure Information to Develop Predictive Models for In Vitro
Toxicokinetic Parameters to Inform High-throughput Risk-assessment. Comput Toxicol. 16.

Rajagopal, R., et al., 2022. Beyond AOPs: A Mechanistic Evaluation of NAMs in DART Testing. Front
Toxicol. 4, 838466.

Ring, C., et al., 2021. Predictive modeling of biological responses in the rat liver using in vitro Tox21
bioactivity: Benefits from high-throughput toxicokinetics. Comput Toxicol. 18.

Ring, C. L., et al., 2019. Consensus Modeling of Median Chemical Intake for the U.S. Population Based on
Predictions of Exposure Pathways. Environ Sci Technol. 53, 719-732.

Rogers, D., Hahn, M., 2010. Extended-connectivity fingerprints. J Chem Inf Model. 50, 742-54.

Sayre, R. R., et al., 2020. Database of pharmacokinetic time-series data and parameters for 144
environmental chemicals. Sci Data. 7, 122.

Shah, I., et al., 2016. Systematically evaluating read-across prediction and performance using a local

validity approach characterized by chemical structure and bioactivity information. Regul Toxicol
Pharmacol. 79, 12-24.

Shah, I., et al., 2021. Generalised Read-Across prediction using genra-py. Bioinformatics.

Sipes, N. S., et al., 2013. Profiling 976 ToxCast chemicals across 331 enzymatic and receptor signaling
assays. Chem Res Toxicol. 26, 878-95.

Sipes, N. S., et al., 2017. An Intuitive Approach for Predicting Potential Human Health Risk with the
Tox2110k Library. Environ Sci Technol. 51, 10786-10796.

Smit, I. A., et al., 2021. Systematic Analysis of Protein Targets Associated with Adverse Events of Drugs
from Clinical Trials and Postmarketing Reports. Chem Res Toxicol. 34, 365-384.

Speen, A. M., et al., 2022. Benchmark Dose Modeling Approaches for Volatile Organic Chemicals Using a
Novel Air-Liquid Interface In Vitro Exposure System. Toxicol Sci. 188, 88-107.

Thomas, R. S., et al., 2019. The Next Generation Blueprint of Computational Toxicology at the U.S.
Environmental Protection Agency. Toxicol Sci. 169, 317-332.

USEPA, 2020a. Chemical and Products Database (CpDat) MySQL Data.
https://doi.org/10.23645/epacomptox.5352997.

USEPA, User's Guide for T.E.S.T version 5.1: Toxicity Estimation Software Tool: A Program to Estimate

Toxicity from Molecular Structure. In: Martin, T. M., (Ed.). Office of Research and Development,
Center for Computational Toxicology and Exposure, Cincinnati, OH, 2020b.

USEPA, National PFAS Testing Strategy: Identification of Candidate Per- and Poly-fluoroalkyl Substances
(PFAS) for Testing. Office of Pollution Prevention and Toxics (OPPT) and Office of Research and
Development (ORD) https://www.epa.gov/svstem/files/documents/2021-10/pfas-natl-test-
strategy.pdf. Washington, DC 20460, 2021a.

USEPA, New Approach Methods Work Plan: Reducing use of animals in chemical testing. Office of
Research and Development and Office of Chemical Safety and Pollution Prevention,
https://www.epa.gov/system/files/documents/2021-ll/nams-work-plan 11 15 21 508-
tagged.pdf, 2021b.

USEPA, A Proof-of-Concept Case Study Integrating Publicly Available Information to Screen Candidates
for Chemical Prioritization under TSCA. https://doi.org/10.23645/epacomptox.14878125, 2021c.

USEPA, Modernizing the Process and Bringing Innovative Science to Evaluate New Chemicals Under
TSCA. Office of Pollution Prevention and Toxics; Office of Research and Development,
https://www.regulations.gov/docket/EPA-HQ-QPPT-2.022-0218/document. 2022a.

USEPA, Multimedia Monitoring Database (MMDB). Office of Research and Development, Center for

Computational Toxicology and Exposure, https://doi.org/10.23645/epacomptox.17065024.vl.
2022b.

Valentin, J. P., et al., 2018. In vitro secondary pharmacological profiling: An IQ-DruSafe industry survey
on current practices. J Pharmacol Toxicol Methods. 93, 7-14.

47


-------
BOSC Review Draft

van Tongeren, T. C. A., et al., 2021. Next generation risk assessment of human exposure to anti-

androgens using newly defined comparator compound values. Toxicol In Vitro. 73, 105132.

Wambaugh, J. F., et al., 2019. New approach methodologies for exposure science. Current Opinion in
Toxicology. 15, 76-92.

Watford, S., et al., 2019. ToxRefDB version 2.0: Improved utility for predictive and retrospective
toxicology analyses. Reprod Toxicol. 89, 145-158.

Williams, A. J., et al., 2017. The CompTox Chemistry Dashboard: a community data resource for
environmental chemistry. J Cheminform. 9, 61.

Willis, C., et al., 2020. Phenotypic Profiling of Reference Chemicals across Biologically Diverse Cell Types
Using the Cell Painting Assay. SLAS Discov. 25, 755-769.

Yang, C., et al., 2015. New publicly available chemical query language, CSRML, to support chemotype
representations for application to data mining and modeling. J Chem Inf Model. 55, 510-28.

Zavala, J., et al., 2017. Regulating temperature and relative humidity in air-liquid interface in vitro

systems eliminates cytotoxicity resulting from control air exposures. Toxicol Res (Camb). 6, 448-
459.

Zavala, J., et al., 2018. A new cell culture exposure system for studying the toxicity of volatile chemicals
at the air-liquid interface. Inhal Toxicol. 30, 169-177.

48


-------