1
QSAR and chemical read-across within the context of
non-testing strategies
Andrew Worth
Institute for Health & Consumer Protection
Joint Research Centre, European Commission
1st SETAC Europe Special Science Symposium: Integrated Testing Strategies for REACH
Brussels, Belgium; 23-24 October 2008
http://ecb.jrc.ec.europa.eu/qsar/qsar-tools/
Overview
2
1. Filling data gaps: categories, read-across and (Q)SARs
2. Reporting formats for read-across and (Q)SARs
3. Non-testing strategy – a stepwise approach
4. Conclusions
Filling data gaps by read-across
3
Known information on the property of a substance (source chemical) is
used to make a prediction of the same property for another substance
(target chemical) that is considered “similar”
Property
Source chemical
Target chemical


1,2-Benzenedicarboxylic acid,
bis(2-ethoxyethyl) ester
diethyl phthalate
Acute fish toxicity?
Known to be harmful: 1 < log LC50 < 2
Predicted to be harmful
Read-across in the analogue approach
4
The analogue approach refers to the grouping of chemicals and application of
read-across for a given endpoint based on a relatively small number of
analogues
Substance 1
one-to-one
Property

Substance 1
many-to-one
Property

Substance 2


reliable data point

missing data point
Substance 2
Substance 3


Read-across in the category approach
5
Chemical 1
Chemical 2
Chemical 3
Chemical 4
Property 1




Property 2




Property 3




Property 4




read-across
Trend analysis
Activity 1




Activity 2




Activity 3




Activity 4




 reliable data point
 missing data point
The category approach refers to a wider approach, based on more analogues,
multiple endpoints, and in which trends are also apparent
Chemical grouping in REACH
6
• Within the analogue and category approaches, data gaps can be filled by
read across
• This avoids the need to test every substance for every endpoint
• Results should be adequate for classification and labelling and/or risk
assessment
• Adequate and reliable documentation of the applied method shall be provided
(Q)SARs in REACH
7
Results obtained from valid qualitative or quantitative structure-activity
relationship models ((Q)SARs) may indicate the presence or absence of a
certain dangerous property. Results of (Q)SARs may be used instead of testing
when the following conditions are met:
• results are derived from a (Q)SAR model whose scientific validity has been
established
• the substance falls within the applicability domain of the (Q)SAR model
• results are adequate for the purpose of classification and labelling and/or risk
assessment, and
• adequate and reliable documentation of the applied method is provided
The Agency in collaboration with the Commission, Member States and interested
parties shall develop and provide guidance in assessing which (Q)SARs will meet
these conditions and provide examples.
Adequacy of (Q)SAR prediction
8
In order for a (Q)SAR result to be adequate
for a given regulatory purpose, the
following conditions must be fulfilled:
•
•
the estimate should be generated by a
valid (reliable) model
Reliable
(Q)SAR
result
(Q)SAR model
applicable to
query chemical
Adequate
(Q)SAR
result
QMRF
the model should be applicable to the
chemical of interest with the necessary
level of reliability
QPRF
•
Scientifically
valid QSAR
model
(Q)SAR model relevant
to regulatory purpose
the model endpoint should be relevant for
the regulatory purpose
QPRF
http://ecb.jrc.ec.europa.eu/qsar/qsar-tools/
Standardised (Q)SAR Reporting Formats
9
The need for “adequate and reliable” documentation is met by using
standardised reporting formats:
A (Q)SAR Model Reporting Format (QMRF) is a robust summary of a (Q)SAR
model, which reports key information on the model according to the OECD
validation principles
A (Q)SAR Prediction Reporting Format (QPRF) is a description and
assessment of the prediction made by given model for a given chemical
Structure
QSAR
Models
methodology
(QMRF)
Prediction
(QPRF)
Biological activity
Validation
Assessment
(Q)SAR Reporting Formats: QMRF
10
QMRF captures information on fulfilment of OECD validation principles, but no
judgement or “validity statement” is included
A (Q)SAR should be associated with the following information:
1.
a defined endpoint
an unambiguous algorithm
3. a defined applicability domain
2.
appropriate measures of goodness-of-fit, robustness and predictivity
5. a mechanistic interpretation, if possible
4.
•
•
•
•
Principles adopted by 37th Joint Meeting of Chemicals Committee and Working
Party on Chemicals, Pesticides & Biotechnology; 17-19 Nov 2004
ECB preliminary Guidance Document published in Nov 2005
OECD Guidance Document published in Feb 2007
OECD Guidance summarised in REACH guidance (IR and CSA)
(Q)SAR Reporting Formats: QPRF
11
QPRF captures information on the substance and its prediction, and is intended
to facilitate considerations of the adequacy of a prediction
1. Substance information
2. General (administrative) information on QPRF
3. Information on prediction (endpoint, algorithm, applicability domain,
uncertainty, mechanism)
4. Adequacy (optional, legislation-specific, and includes judgement and
indicates whether additional information is needed for WoE assessment)
Adequacy assessment is a sensitive issue:
• Depends on reliability and relevance of prediction, but also on the availability
of other information, and the consequence of being wrong. Not just a scientific
consideration, but also a policy decision
• Development of “Klimisch-type” codes for predictions not supported by QSAR
Working Group
Analogue Reporting Format
12
Documents the outcome of applying the analogue approach
1.
2.
3.
4.
5.
6.
Hypothesis for the analogue approach
Source chemical(s)
Purity / Impurity profiles for source and target chemicals
Justification of analogue approach (based on available experimental data)
Data matrix (for source and target chemicals)
Regulatory conclusions on adequacy (for REACH):
- C&L is similar to the source chemical
- PBT / vPvB assessment is similar to the source chemical
- for risk assessment, the dose descriptor is similar to the source chemical, or
adaptations are necessary
- uncertainties that need to be addressed
Outline of a non-testing strategy
13
Existing information
Preliminary assessment of reactivity and fate
Working Matrix
A B C
Chemical
Metabolite 1
Classification schemes and structural alerts
Metabolite 2
Preliminary assessment of reactivity, fate & toxicity
Assess adequacy
Read-across
(Q)SARs
Conclusion or
targeted testing
Step 1: Information collection
14
• Chemical composition (components, purity/impurity profile)
• Structure generation and verification
• Key chemical features (functional groups, protonation states, isomers)
• Experimental data: physicochemical properties, (eco)toxicity, fate
• Freely-accessible web resources (ESIS, ChemSpider, PubChem, AMBIT2)
• Databases in freely-available software tools (OECD Toolbox)
• Commercial databases (Vitic, …)
• Estimated data: pre-generated QSAR or read-across estimates
• Freely-accessible web resources (ChemSpider, Danish QSAR database)
• Chemical category databases (OECD Toolbox)
European chemical Substances Information System
15
ESIS
http://ecb.jrc.ec.europa.eu/esis/
ENDDA – Endocrine Active Chemicals Database
16
web-accessible database under development
Step 2: Preliminary assessment of reactivity & fate
17
• Prediction of abiotic / biotic reactivity to identify reactive potential and
possible transformation products / metabolites
• Freely-available software
• CRAFT (Chemical Reactivity & Fate Tool)
• START (Structural Alerts in Toxtree)
• OECD Toolbox
• Commercial software and databases
• CATABOL, TIMES, Meteor, Mexalert, MetabolExpert …
• MetaPath, SciFinder, MDL Reaction Database …
START - STructural Alerts for Reactivity in Toxtree
18
• Collaboration with Molecular Networks (Germany)
• Toxtree plug-in
• Estimates biodegradation potential
CRAFT - Chemical Reactivity And Fate Tool
19
• Collaboration with Molecular Networks (Germany)
• Generates & visualises reactions, ranks transformation products
• Initial emphasis on abiotic processes & microbial biodegradation
• Data model based on AMBIT technology
• User can modify knowledge base and rulebase
Steps 3 & 6: SARs, QSARs and expert systems
20
• Models and rulebases for mode-of-action classification, hazard identification,
hazard classification and potency prediction
• Freely-available software
• Episuite, Toxtree, AMBIT2, OECD Toolbox …
• OpenTox framework (http://www.opentox.org)
• Commercial software
• DEREK, MultiCASE, HazardExpert, ToxAlert, ToxBoxes …
• Insilicofirst consortium (Multicase Inc, Lhasa Ltd, Molecular Networks GmbH,
Leadscope Inc)
• QSAR Model Databases (QMDBs)
• JRC QSAR Model Database
• OECD Toolbox
Toxicity Estimation Tool: Toxtree
21
Toxtree is a flexible, user-friendly, open source application, which is able to
classify chemicals into modes of action and estimate toxic hazard by
applying decision tree approaches
Collaboration with Ideaconsult (BG)
Rulebases in version 1.51 (June 2008):
• Acute Fish Toxicity (Verhaar scheme)
• Oral systemic toxicity (Cramer scheme)
• Skin irritation & corrosion potential (BfR rulebase)
• Eye irritation & corrosion potential (BfR rulebase)
• Mutagenicity & carcinogenicity (Benigni-Bossa rulebase)
http://ecb.jrc.ec.europa.eu/qsar/qsar-tools/
Main screen in Toxtree
22
Compound properties
Prediction
Compound structure
Reasoning
Biinary decision trees in Toxtree
23
Patlewicz G, Jeliazkova N, Safford RJ, Worth AP & Aleksiev B (2008). An evaluation of the
implementation of the Cramer classification scheme in the Toxtree software. SAR and QSAR in
Environmental Research 19, in press.
Benigni-Bossa mutagenicity rules in Toxtree
24
SARs
• Ashby, 1985
(Ashby’s “poly-carcinogen”)
• Ashby and Tennant, 1998
• Kazius et al., 2005
• Kazius et al., 2006
• Bailey et al., 2005
QSARs derived by discriminant analysis
Probability
of effect
= F
Descriptors:
Hydrophobic
log P
+
Electronic
HOMO, LUMO
+
Steric
MR
Step 5: Chemical grouping and read-across
25
• Chemical read-across within analogue and category approaches
• Biological read-across (between endpoints or species)
• Chemical grouping by a top-down approach
Inventory / dataset
• Supervised and unsupervised statistical methods
• Ranking methods (DART)
Chemical group
• Chemical grouping by a bottom-up approach
Source chemical
• Freely available tools with analogue-searching capability
(Toxmatch, AMBIT2, AIM, PubChem, OECD Toolbox)
Worth A et al (2007). The Use of Computational Methods in the Grouping
and Assessment of Chemicals - Preliminary Investigations. EUR 22941 EN
Chemical group
Ranking tool: DART
26
DART (Decision Analysis by Ranking Techniques) is a flexible, user-friendly,
open source application, which is able to rank and group chemicals
according to properties of concern
hazard
a
b
e
Level 3
a, b, e:
d: mini
c
Level 2
a, b, e:
a, c, d:
d
Level 1
• collaboration with Talete srl (Italy)
• supports priority setting of chemicals
http://ecb.jrc.ec.europa.eu/qsar/qsar-tools/
b, c, d:
Toxmatch: chemical similarity tool
27
Collaboration with Ideaconsult (BG)
http://ecb.jrc.ec.europa.eu/qsar/qsar-tools/
Supports:
• chemical grouping & read-across
• comparison of training & test sets
Read-across in Toxmatch
28
Many-to-one read-across of a quantitative property (k Nearest Neighbours)
Patlewicz G, Jeliazkova N, Gallegos Saliner A & Worth AP (2008). Toxmatch – A new software tool to aid
in the development and evaluation of chemically similar groups. SAR and QSAR in Environmental
Research 19, 397-412.
Example of read-across in Toxmatch
29
• BCF of aniline predicted on basis of similarity to training set
• Similarity defined by 3 descriptors: effective diameter, maximum diameter and LogP
Aniline test chemical
Pairwise similarity
between aniline
and training set
compounds
Example of read-across in Toxmatch
30
NH2
H2C
N
Similarity index: 0.022
LogBCF (est) = 1.4306
LogBCF = 1.83
Predicted LogBCF of
aniline is 1.05
Experimental LogBCF = 0.78
(Hazardous Substances Databank)
NH CH3
Similarity index: 0.025
LogBCF (est) = 0.856
LogBCF = 0.41
Step 6: JRC QSAR Model Database
31
Under REACH, there is a need to identify reliable and relevant (Q)SARs and to
use them to derive estimates and/or have access to their pre-calculated
estimates
The JRC QSAR Model Database is an inventory of information on (Q)SAR
models
• collaboration with Ideaconsult (BG)
• based on the AMBIT software for chemoinformatics data management
• developers and users of (Q)SAR models can submit information on (Q)SARs by
using the (Q)SAR Model Reporting Format (QMRF)
• the QMRF is a communication tool between industry and the authorities under
REACH
• inclusion of the model in the QSAR Model Database does not imply acceptance or
endorsement by the JRC or the European Commission
JRC QSAR Model Database
32
UPLOAD
QMRF (xml)
http://qsardb.jrc.it
Access through Internet
Upload of QMRF in xml and sdf formats
Upload of training & test sets in sdf format
DOWNLOAD
QMRF
Pdf report
QMRF
Excel file
Generation of QMRF from scratch
Download of QMRF in various formats
QMRF (sdf)
QMRF
Xml file
Automatic submission of QMRF to JRC
Ability to search QSAR database
QMRF
Sdf file
Submission, review & publication process
33
The submission, review and publication of QMRFs mimics the process in peerreview scientific journals
Anonymous users can search for documents
Four role types can be assigned to registered users:
Author, Reviewer, Editor and Administrator
• The Author can create and submit a QMRF document. Once submitted, the
documents are reviewed by a Reviewer
• The Reviewer can recommend publication of the document, or return it to
the Author, if there is insufficient/unclear information
• The Editor assigns Reviewers to the newly submitted documents.
• The Administrator manages the system
Searching the database (1)
34
Anonymous users can search by document or by substance
QMRF document searches can apply the following criteria (joined by AND or
OR operators):
•
•
•
•
•
•
QMRF No.
Free text
Endpoint
Algorithm
Software
Authors
Searching the database (2)
35
Structure searching can be based on exact structure or similarity
•
•
•
•
•
CAS No.
Formula
Chemical name
Alias
SMILES
Concluding remarks
36
• To optimise the use of non-testing data, an integrated approach for the use
of multiple models and tools is required
• A conceptual framework for integrating non-testing data is provided in the
REACH guidance document
• An increasingly wide range of models are being implemented in software
tools
• There is a need to facilitate the use of multiple tools by developing
automated workflows
• Further guidance is needed on how to assess the adequacy of non-testing
data by weight-of-evidence approaches