Hi folks,

An interesting report on the use of alternatives to testing on animals
for the REACH regulation has been published by ECHA on July 29th,
2011. It can be downloaded from:

http://echa.europa.eu/doc/117reports/alternatives_test_animals_2011_en.pdf

Here are some important quotes and some further comments (prefixed
with VJ: and enclosed in square brackets):

-- p. 14: "There are many research projects ongoing within and beyond
the EU that focus on replacement, reduction or refinement of animal
testing. Important examples within the EU are research initiatives
that have been launched in recent last years. These include the
development and optimisation of reproductive toxicology (Re-Pro-Tect;
2004-2009), acute toxicity (A-Cute-Tox; 2004-2009), skin and
respiratory sensitisation (sens-it-iv; 2005-2010), carcinogenicity
(Carcinogenomics; 2006-2011), chronic toxicity (Predict-iv; 2008-2013)
and repeated dose toxicity (COLIPA-DG RTD Joint Research Initiative;
2009). In the future, these research projects may deliver in the
future (sic) new approaches to combine different tests in the most
optimal way (testing strategies) for these endpoints. For more
details, please visit the ECVAM website." [VJ: note that neither
OpenTox, nor Cadaster are mentioned here; looks like we need to do
more to increase our visibility on ECHA's radar screen; I think that
this issue could be addressed best by the project coordinators in
cooperation with the scientific project officers]

-- p. 15: "It has to be noted that the results of research projects
and new or refined methods based on those results can be used for
regulatory purposes once they are validated and adopted. In the area
of alternative methods to animal testing, such validations are
conducted according to internationally agreed principles. In Europe,
ECVAM is the responsible body for the validation of these methods, and
more internationally, the OECD plays a role. This process is rather
time consuming and may take years for a specific method to be
internationally validated and accepted for regulatory purposes." [VJ:
note that validation has a substantially different meaning in machine
learning and regulatory context (I intentionally avoid the discussion
here whether this is sound or not...); OpenTox addresses quite well
the validation in the machine learning sense, however AFAIK we haven't
done much yet on the ECVAM/OECD front -- again, this could be best
tackled by the project coordinator in cooperation with the scientific
project officer and authorised representatives of ECVAM/OECD. Last but
not least, ECHA acknowledges here that the "process is rather time
consuming and may take years for a specific method to be
internationally validated and accepted for regulatory purposes". This
should answer any concerns or criticisms regarding whether OpenTox
models have been validated and accepted for regulatory purposes -- in
the light of ECHA's standpoint on this, such validation is clearly
beyond the timeframe of the project and the competences/authority of
the project partners]

-- p. 16: "The OECD QSAR Toolbox is an important tool for supporting
and enabling category building. ECHA actively contributes to the
further development of this Toolbox. This freely available software is
useful to group chemicals and apply read-across techniques for
assessing the (eco)toxicity hazards of chemical substances under
REACH." [VJ: would be nice to see ECHA endorse the OpenTox framework
at least to the same level as the OECD QSAR Toolbox, but I might be
asking for too much...again something that is probably achievable by
some joint efforts of the project coordinator and the scientific
officer]

-- p. 16: "ECHA is collaborating with the JRC Computational Toxicology
Group in its mission to promote the availability for regulatory use of
valid computer-based methods, for example, QSARs, used for assessing
the intrinsic properties of chemical substances. More information,
including structured and peer-reviewed documentation of (Q)SAR models
and free access to JRC QSAR Model Database is available from the JRC
website." [VJ: would be nice to submit and see published some of the
relevant OpenTox models in the QMRF inventory, hosted by JRC and
available at http://qsardb.jrc.it/qmrf/]

-- p. 16: "There are a number of promotion platforms and organisations
which actively contribute to the promotion and/or development of
alternative methods, e.g. the European Partnership for Alternative
Approaches to Animal Testing (EPAA), European Consensus-Platform for
Alternatives (ECOPA), and many others. The International Council on
Animal Protection in OECD Programmes, ICAPO is an international
organisation that develops guidelines and programmes for the testing
of chemicals. ICAPO is currently working with the Working Group of
National Co-ordinators of the Test Guideline Program, Task Forces on
Endocrine Disruptor Testing and and Existing Chemicals, including
(Q)SAR, and in the Validation Management Groups for Non-Animal testing
and for Mammalian testing. For the up-to-date progress in the
development of alternatives to animal testing the reader is referred
to the above listed organisations and their websites, which provide
useful information." [VJ: Might be a good idea to work on increasing
awarness of OpenTox results within those (and similar) organizations]

-- p. 26: [VJ: Distribution of data sources for Acute Toxicity; read
accros 21.4%, QSAR 0.1% of the ESRs]

-- p. 27: [VJ: Distribution of data sources for Skin irritation in
vitro; read across 11.9%, QSAR 0.0% of the ESRs]

-- p. 28: [VJ: Distribution of data sources for Skin irritation in
vivo; read across 21.3%, QSAR 0.1% of the ESRs]

-- p. 29: [VJ: Distribution of data sources for Eye irritation in
vitro; read across 7%, QSAR 0% of the ESRs]

-- p. 30: [VJ: Distribution of data sources for Eye irritation in
vivo; read across 20.9%, QSAR 0% of the ESRs]

-- p. 31: [VJ: Distribution of data sources for Skin sensitisation in
vivo; read across 20.8%, QSAR 0.5% of the ESRs]

-- p. 33: [VJ: Distribution of data sources for Repeated Dose toxicity
(all routes and all durations); read across 28.1%, QSAR 0.1% of the
ESRs]

-- p. 35: [VJ: Distribution of data sources for Genetic toxicity in
vitro; read across 22%, QSAR reported as 0% of the ESRs, but obviously
a typo/rounding error, since in fact 5 ESRs were based on QSARs
according to the same table]

-- p. 36: [VJ: Distribution of data sources for Genetic toxicity in
vivo; read across 24.8%, QSAR 0% of the ESRs]

-- p. 38: [VJ: Distribution of data sources for Toxicity to
reproduction; read across 23.8%, QSAR 0.1% of the ESRs]

-- p. 39: [VJ: Distribution of data sources for Developmental
toxicity; read across 29.7%, QSAR 0.2% of the ESRs]

-- p. 40: [VJ: Distribution of data sources for Carcinogenicity; read
across 27.9%, QSAR 0.2% of the ESRs]

-- p. 42 : [VJ: Distribution of data sources for Bioaccumulation in
fish; read across 24.7%, QSAR 3.1% of the ESRs]

-- p. 43: [VJ: Distribution of data sources for Toxicity to fish; read
across 20.2%, QSAR 2.1% of the ESRs]

-- p. 44: [VJ: Distribution of data sources for Long term toxicity to
fish; read across 21.2%, QSAR 4.3% of the ESRs]

-- p. 45: [VJ: Distribution of data sources for Long term toxicity to
birds; read across 6.4%, QSAR reported as 0% of the ESRs, but
obviously a typo/rounding error, since in fact 1 ESR was based on
QSARs according to the same table]

-- p. 54: "When interpreting the findings of the ESR analysis, it
should be noted that in principle there may be deficiencies discovered
in the compliance check dossier evaluation work that result in further
animal studies being requested if the quality of either the
experimental data or the justifications for the adaptations in the
dossiers is discovered to be inadequate. This was found for
read-across approaches as well as for the options to omit the study.
These two approaches are identified in the current report as the main
options used by the registrants for higher tier tests if they did not
use experimental data or submitted a testing proposal." [VJ: looks
like read across is more likely to *delay* experimental tests through
administrative ping-pong plays, rather than completely avoid animal
testing in most of the cases]

As you can see from the above statistics, QSARs are rarely used (if at
all) in endpoint study records. One reason for this might be that
published/validated models are usually trained on public data, while
in most cases companies have models trained on proprietary data that
capture somehow better the specifics of the chemicals they're dealing
with. It might be economically more advantageous to apply read-across
as a first step and consider other alternative methods or experiments
only if/when required by ECHA. It is also worth mentioning that the
OECD QSAR toolbox provides relatively good support for read-across. In
this context it is important to keep in mind that the low usage of
QSARs for REACH can hardly be attributed as a failure of a single
project like OpenTox, but is rather a much more generic issue which
has probably some scientific, regulatory, economic and cultural
aspects.

Kind regards,
Vedrin