Read-across, a popular data gap filling technique, traditionally relies on a thorough expert assessment. This approach can lead to inconsistent predictions, has limitations in terms of the numbers of chemicals that can be evaluated and offers little insight into the generalizability of the approach or its performance. We sought to evaluate the baseline performance of read-across for a large set of chemicals in a systematic, objective and reproducible manner and to provide quantitative measures of uncertainty for the predictions derived. The approach developed, generalized read-across (GenRA), relies on chemical descriptor information and/or in vitro bioactivity data (from ToxCast high throughput screening data) to derive read-across predictions of effects observed in in vivo repeat-dose toxicity studies. A web-based GenRA tool was then developed anchored around the established category workflow and a dynamic grid interface structure. The default starting point is identifying source analogues with associated in vivo data on the basis of chemical fingerprints. The next step is to analyze the scope and quantity of available data (both in vitro and in vivo). The third step generates a data matrix in order to evaluate the analogues – in terms of their consistency and concordance of effects across the different toxicity effects. The final steps involve generating a GenRA prediction and exporting the predictions. Here we describe the functionality and features of the GenRA web application which has recently been released as a new feature in the US EPA CompTox Chemistry Dashboard. Users are able to search the dashboard for a substance of interest, browse available information including toxicity information and then derive GenRA predictions. GenRA offers a novel and practical means of being able to perform objective read-across that can be helpful in screening level hazard assessments. This abstract does not necessarily represent U.S. EPA policy.

1. Translating research into practical tools:
A case study of GenRA,
a new read-across tool
Antony Williams1, George Helman2, Jeff Edwards1, Jeremy Dunne1,
Imran Shah1 and Grace Patlewicz1
1) National Center for Computational Toxicology, U.S. Environmental Protection Agency, RTP, NC
2) Oak Ridge Institute of Science and Education (ORISE) Research Participant, RTP, NC
August 2018
ACS Fall Meeting, Boston
http://www.orcid.org/0000-0002-2668-4821
The views expressed in this presentation are those of the author and do not necessarily reflect the views or policies of the U.S. EPA

2. • National Center for Computational Toxicology
established in 2005 to integrate:
– High-throughput and high-content technologies
– Modern molecular biology
– Data mining and statistical modeling
– Computational biology and chemistry
• Researching computational approaches to
quickly evaluate the safety of chemicals for
potential risk.
• Outputs: a lot of data, models, algorithms and
software applications
National Center for
Computational Toxicology

3. The CompTox Dashboard
https://comptox.epa.gov/dashboard
• A publicly accessible website delivering access:
– ~762,000 chemicals with related property data
– Searchable by chemical, product use, gene and assay
– Experimental and predicted physicochemical property
data, environmental fate and transport, and tox endpoints
– “Bioactivity data” for the ToxCast/Tox21 project – plus
derived models
– NEW Generalized Read-Across (GenRA) module
– “Batch searching” of predicted data for 1000s of chemicals
2

4. CompTox Dashboard
https://comptox.epa.gov/dashboard
3

5. CompTox Dashboard
Chemicals
4

6. Detailed Chemical Pages
5

7. Physicochemical properties
6

8. Hazard Data – Human and Eco
7

9. Bioactivity Data (ToxCast/Tox21)
Data below for Bisphenol A
8

10. Other Dashboard Predictions
• Predictions and models expand outside of
simply physicochemical and environmental
fate and transport
• Examples
– Read-across for Toxicity Endpoints
– Quantitative Structure–Use Relationship (QSUR) models
– High-Throughput ToxicoKinetics (HTTK)
– Models based on high throughput bioactivity data
9

11. Real-Time Predictions
10

12. Real-Time Predictions
11

13. Real-Time Predictions
12

14. Definitions: Read-Across
• Known information on the property of a substance
(source) is used to make a prediction of the same
property for another substance (target) that is
considered “similar”
13
Source chemical Target chemical
Property  


Reliable data
Missing data
Predicted to be harmfulKnown to be harmful
Acute fish toxicity?

15. GenRA (Generalised Read-Across)
• Predicting toxicity as a similarity-weighted activity
of nearest neighbors based on chemistry and/or
bioactivity descriptors
• Goal: to systematically evaluate read-across
performance and uncertainty using available data
• The approach enabled a performance baseline
for read-across predictions of toxicity effects
within specific study outcomes to be established
14

16. Read-across workflow in GenRA
15
Decision
Context
Screening level assessment
of hazard based on
toxicity effects from
ToxRefDB
Analogue
identification
Similarity context is based
on structural
characteristics
Data gap
analysis for
target and
source
analogues
Analogue
evaluation
Evaluate consistency and
concordance of
experimental data of
source analogues across
and between endpoints
Read-across
Similarity weighted
average – many to one
read-across
Uncertainty
assessment
Assess prediction and
uncertainty using AUC and
p value metrics

17. GenRA (Generalised Read-Across)
16

18. GenRA (Generalised Read-Across)
Structure Similarity
Select and Review Analogs

19. GenRA (Generalised Read-Across)
Review Available Data Fingerprint indicating available dataSelect and Review Analogs

20. GenRA (Generalised Read-Across)
19
Run GenRA
Target
Source analogues
Red : Toxicity effects.
Blue: No Toxicity effects
Grey : Absence of data

21. Demonstration

22. GenRA is one of multiple
Read-Across Tools available
21
Tool AIM ToxMatch AMBIT OECD
Toolbox
CBRA ToxRead GenRA
Analogue
identification
X X X X X X X
Analogue
Evaluation
NA X X
by other
tools
available
X X X
For
Ames &
BCF
NA
Data gap
analysis
NA X X
Data
matrix
can be
exported
X
Data
matrix
viewable
NA NA X
Data matrix
can be
exported
Data gap filling NA X User
driven
X X X X
Uncertainty
assessment
NA NA NA X NA NA X
Availability Free Free Free Free Free Free Free

23. Related Publications
22

24. Conclusions
• The CompTox Dashboard delivers experimental
and predicted data for physchem, environ. fate
and transport
• A new Read-Across module, GenRA, is now
available
• Real time predictions are also possible –
coming soon pKa and logD predictions
23

25. Acknowledgments
National Center Comp. Tox.
• Imran Shah
• George Helman
• Prachi Pradeep
• Tony Williams
• Jeff Edwards
• Jeremy Dunne
• NCCT Development team
• Chris Grulke
• Reeder Sams
• Katie-Paul Friedman
• Rusty Thomas
24
National Center for
Environ. Assessment
• Jason Lambert
• Lucy Lizarraga
• Mark Cronin LJMU

26. Contact
Antony Williams
US EPA Office of Research and Development
National Center for Computational Toxicology (NCCT)
Williams.Antony@epa.gov
ORCID: https://orcid.org/0000-0002-2668-4821
25