The Toxicologist - Society of Toxicology

suggest that the combination of tools to address reactive metabolite risk can provide
design teams with an assessment of real risk rather than the simple hazard alert of a
structural alert.
155 DESIGN AND DEVELOPMENT OF AN
INSTITUTIONAL KNOWLEDGE-BASE AT FDA’S
CENTER FOR FOOD SAFETY AND APPLIED
NUTRITION.
K. Arvidson, A. McCarthy, C. Yang and D. Hristozov. CFSAN, U.S. FDA,
College Park, MD.
The Chemical Evaluation and Risk Estimation System (CERES) project under development
in FDA’s Center for Food Safety and Applied Nutrition aims at establishing
a sustainable data management and storage system that will provide decision
support tools for both pre-market and post-market safety assessments of food additives
and food contact substances as well as for potential contamination issues. The
development of CERES will provide a single unified data repository that compiles
available information on a substance, including: chemical structures and properties,
regulation records, toxicity studies, and other biological screening assays. In cases
where no information is available for a particular substance, CERES provides tools
to identify potential safety concerns by applying mode of action-driven QSAR prediction
models as well as to identify and analyze data on structural and biological
analogs (read-across). To achieve this goal, CERES requires a solid foundation for
handling chemical and toxicity data as well as knowledge (e.g., rules or computational
toxicology models) derived from these data. Thus the CERES project not
only needs to address a wide variety of issues in database and knowledgebase construction,
but also in the development of computational methods. Strategies are
presented for tasks of data harvesting, data modeling, development of rule-base and
computational models. This poster will focus on the overall strategy for the CERES
knowledgebase; each sub-area will also be presented in other posters.
156 PHARMACOLOGICAL PROFILE SIMILARITY – A
RELEVANT METHOD FOR ANTICIPATION OF IN VIVO
SAFETY?
D. Muthas 1 , M. Rolf 2 , S. Matis-Mitchell 3 and S. Boyer 1 . 1 Global Safety
Assessment, AstraZeneca R&D, Mölndal, Sweden, 2 Global Safety Assessment,
AstraZeneca R&D, Alderly Park, United Kingdom and 3 Discovery Information,
AstraZeneca R&D, Wilmington, DE.
We have analyzed a pharmacological profile of historical AstraZeneca development
compounds in order to investigate their potential to predict in vivo findings. A
panel of 100 targets, most of which have been chosen for their association with adverse
effects, was used to derive a pharmacological ‘biospectrum’ of each compound.
The similarities of these biospectra were then calculated using several different
similarity metrics (e.g. Tanimoto, Dice, Yule, Euclidian distance and
Mahalanobis distance) and activity binning to identify compounds sharing similar
biological fingerprints. To evaluate their relevance, these similarities were then compared
to the preclinical and/or clinical findings of the compounds found by mining
preclinical study reports and clinical safety databases. The observations were later
analyzed using all found data as well as with special focus on certain types of studies
(eg 1 month dog and 2 week rat). The compounds were then clustered based in
the in vivo effects and compared to the clusters based on biospectrum. The results
of these studies indicate that several pairs of structurally dissimilar compounds with
similar biospectra can be identified and an overlap of their preclinical findings was
frequently found. The frequency of association of the biospectrum and outcome
was assessed for statistical significance and found to be significant in several cases.
This suggests that pharmacological profiling and establishing a biospectrum could
help identify the potential for safety issues from more than the chemical structure
or the activity on a single target.
157 DEVELOPMENT OF QSAR MODEL FOR HIGH-SPEED
IN SILICO IDENTIFICATION OF POTENTIALLY
PHOTOTOXIC ORGANIC COMPOUNDS.
W. Plonka 1 and J. M. Ciloy 2 . 1 Life Science, FQS Poland, Cracow, Poland and 2 Life
Science, Fujitsu Kyushu Systems Ltd., Fukuoka, Japan. Sponsor: K. Hayamizu.
Light-exposure induced toxicity to humans is one of factors limiting the use of
chemical compounds in wide range of applications, starting with cosmetics, household
chemistry products, as well as pharmaceutical industry. The possibility of reliable
in-silico identification of potentially phototoxic compounds would contribute
to product safety and reduce development costs. A Qualitative Structure-Activity
Relationship study had been done on a set of 229 structures, of which 114 were
known from public literature to cause phototoxic effects, and the rest were neutral
compounds, including compounds routinely used in commercial cosmetics. The
objective was to create a model that would have as low as possible false-negative
classification rate to ensure high safety margin of the predictions. It was assumed,
that a molecule has to satisfy a set of quantum, as well as structural properties – to
which both light absorption and energy transfer in biological systems relate in order
to be potentially phototoxic. A stochastic gradient perceptron model was created
employing quantum descriptors calculated by fast AM1 semiempirical method,
fragment count and topological descriptors. The model has a 0% false-negative
classification rate on the training set. The overall internal leave-1-out cross validation
rate of the model is over 90%. The details of model building technique and
compounds used in research are presented in the poster.
158 THE POWER OF AN ONTOLOGY-DRIVEN
DEVELOPMENTAL TOXICITY DATABASE FOR DATA
MINING AND COMPUTATIONAL MODELING.
A. Richard 1 , E. Julien 2 , E. Busta 3 , M. Wolf 4 and C. Yang 3 . 1 NCCT (D343-03),
U.S. EPA, Research Triangle Park, NC, 2 LifeSci Research Service, Rockville, MD,
3 HFS-2775, U.S. FDA CFSA, College Park, Maryland, MD and 4 Lockheed-Martin,
U.S. EPA, Research Triangle Park, NC.
Modeling of developmental toxicology presents a significant challenge to computational
toxicology due to endpoint complexity and lack of data coverage. These challenges
largely account for the relatively few modeling successes using the structure–activity
relationship (SAR) paradigm. The development of new in vitro
profiling approaches, employing screening assays or lower organisms to evaluate developmental
toxicity requires anchoring to the results of in vivo studies. The
International Life Science Institute (ILSI) recently released a public ontologydriven
DevToxDB with data extracted from published studies. This project heavily
influenced other efforts, and the data content in the ILSI DevToxDB was combined
into an ontology-based database along with other public data from EPA
ToxRefDB, FDA Center for Food Safety and Applied Nutrition and Center for
Drug Evaluation Research and the National Toxicology Program. All chemical
structures are indexed in DSSTox, providing the ability to assess chemical coverage
and diversity of these largely non-overlapping data inventories, which include
drugs, pesticides, industrial chemicals, food ingredients and contact substances.
The use of a common toxicological ontology provides a logical means to group and
aggregate biological effects in subsequent computational analyses.
Chemoinformatics methods are used to compare the chemical space of each data
source, and to explore associations with biological endpoints through the toxicology
ontology perspective. Chemicals in these distinct datasets cause common phenotypes
as well as distinct, non-overlapping phenotypes in the same target organs.
Incorporation of the ILSI DevTox database into other public database efforts
should provide a rich foundation for spurring innovation in SAR modeling of developmental
endpoints. This abstract does not necessarily represent policies of FDA
or EPA.
159 USE OF C. ELEGANS GROWTH ASSAY DATA IN NOVEL
TWO-STEP HIERARCHICAL QSAR MODELING
WORKFLOW ENHANCES IN VIVO TOXICITY
PREDICTION FOR TOXCAST PHASE I CHEMICALS.
A. Golbraikh 1, 2 , R. Shah 1, 4 , W. Boyd 3 , M. Smith 4 , A. Tropsha 1, 2 and J.
Freedman 3 . 1 SciOme LLC, Research Triangle Park, NC, 2 University of North
Carolina at Chapel Hill, Chapel Hill, NC, 3 National Institute of Environmental
Health Science, Research Triangle Park, NC and 4 SRA International, Durham, NC.
The US EPA ToxCast Phase I program provides data for 320 substances with
known in vivo toxicity measured in 76 assays; the results of ca. 600 in vitro assays
for the same substances are available as well. The latter data have been used previously
with varying degree of success to improve the predictive power of in silico
models of chemical toxicity. Herein, we have explored, in the same context, new
whole organism toxicity data generated for ToxCast chemicals in the C. elegans
growth assay. The goal of this study was to establish both the absolute as well as relative
(i.e., with respect to other short term assays included in the ToxCast Phase I
database) value of C. elegans growth assay for predicting chemical in vivo toxicity
and prioritizing new chemicals for in vivo studies. Unlike most in vitro assays where
the number of active compounds was typically lower than that of inactive ones, the
C. elegans assay resulted in nearly balanced dataset. kNN QSAR modeling of C. elegans
data using our standard computational workflow with the emphasis on external
validation resulted in models with the total accuracy of 66%. The C. elegans
SOT 2011 ANNUAL MEETING 33