|
|
Toxicoinformatics Home |
Endocrine Disruptor Knowledge Base
The Endocrine Disruptor Knowledge Base (EDKB)
consists of a number of scientific resources, including a biological
activity database, QSAR training sets and computational models to
predict estrogen and androgen activity. As a knowledge base, it is
intended to serve as a resource for research and regulatory scientists,
and other interested parties, to make use of an existing body of
knowledge, foster the development of computational predictive toxicology
models and reduce dependency upon slow and expensive animal experiments.
Endocrine disruptors are chemicals that interfere with the endocrine
systems, leading to adverse effects. Some chemicals do this by binding
to receptors, such as the estrogen and androgen receptors. The EDKB
website provides access to a relational database containing in vitro and
in vivo experimental data for more than 3000 chemicals, and includes
literature citation and the ability to conduct chemical structure
search. Among the data are two datasets intended for use in predictive
model development. The datasets were derived from competitive binding
assays for the estrogen and androgen nuclear receptor proteins that were
validated and carried out at the NCTR. For both designed training
datasets, chemicals were selected to encompass broad chemical structure
diversity as well as activity range.
A major element of the EDKB program has been the development of
computer-based predictive models to predict affinity for binding of
compounds to the estrogen and androgen nuclear receptor proteins. EDKB
models have been developed using both commercial and NCTR-developed SAR
and QSAR methods. We are currently developing other methods and models
that will enable activity predictions solely based on chemical structure
to be done online. Specifically, we plan to provide online access to a
model based on the
Decision Forest classification method where the user is only
required to input the structure of the chemical to be predicted.
For more in depth review of the EDKB program, the reader is referred
to the program’s
publications.
back to top
This website provides a number of resources for scientists. Click
on the links below to view the desired resource:
EDKB Database
with Chemical Structure Search
Estrogen Receptor Binding
Dataset
Androgen Receptor Binding
Dataset
EDKB Database with Estrogen Activity Prediction of Industrial and
Environmental Chemicals
Comparative Molecular Field Analysis Model for Estrogen Receptor Binding
Comparative Molecular Field Analysis Model for Androgen Receptor Binding
Decision
Forest Model with Prediction Confidence
Decision Forest Predictions for 6500 Industrial and
Environmental Chemicals
Four-Phase
Screening and Priority Setting Model
Future Web-Based Prediction
back to top
- EDKB Database with Chemical Structure Search
The EDKB
database (Tong, 2002) is a curated database containing the EDKB ER and
AR training datasets together with considerable additional data from
the literature for various types of in vitro and in vivo assays. Data
for more than 3200 chemicals; some 2000 relevant citations are
available. The relational database provides Boolean and chemical
structure search, graphical and table displays, as well other
capabilities and data export functions. The EDKB database can be
accessed by clicking the link below:
EDKB Database
back to resources
- Estrogen Receptor Binding Dataset.
Blair (2000) and Branham
(2002) published the EDKB estrogen receptor (ER) binding dataset that
was produced expressly as a training set designed for developing
predictive models. The data is based on a validated assay using rat
uteri. The dataset contains 131 ER binders and 101 non-ER binders.
This structurally diverse data set has 312 predictors generated using
the Molconn-Z software 4.07 and was analyzed using CERP (Ahn et al.,
2007). These training set chemicals were selected for both chemical structure
diversity and range of activity, both of which are essential to
develop robust QSAR and other models (Perkins,
2003). Guided by the
SAR studies described by Fang (2001), the chemicals were selected to
provide uniform coverage of a diverse chemical structure domain, as
well coverage of an activity range extending down a million-fold below
that of the endogenous hormones.
This dataset is available either in
the EDKB Database, or by clicking the
appropriate link below:
Download ER Binding Dataset:
Binding Data (Excel file)
Descriptor Data (Text File)
SD File
back to resources
- Androgen Receptor Binding Data set.
Fang (2003)
published the EDKB androgen receptor (AR) binding dataset that was
produced expressly as a training set designed for developing
predictive models. The data is based on a validated assay using
recombinant AR. The dataset contains 146 AR binders and 56 non-AR
binders. These training set chemicals were selected for both chemical
structure diversity and range of activity, both of which are essential
to develop robust QSAR and other models (Perkins,
2003). Guided by the
SAR studies described by Fang (2003), the chemicals were selected to
provide uniform coverage of a diverse chemical structure domain, as
well coverage of an activity range extending down a million-fold below
that of the endogenous hormones.
This dataset is available either in
the EDKB Database or by clicking
appropriate link below:
Download AR Binding Dataset: Excel
File
Text File
SD File
back to resources
-
Comparative Molecular Field Analysis Model for Estrogen Receptor Binding
Shi (2001) published the CoMFA QSAR model for ER binding that is based on the
EDKB estrogen receptor binding dataset. The model has a cross-validated r2 = 0.66.
back to resources
-
Comparative Molecular Field Analysis Model for Androgen
Receptor Binding Hong (2001) published the CoMFA QSAR model for AR
binding that is based on the EDKB androgen receptor binding dataset.
The model has a cross-validated r2 = 0.57.
back to resources
- Decision Forest Model with Prediction Confidence More
recent EDKB research has focused on developing predictive models that
provide predictions with quantified accuracy. Tong (2004) published a
model based on the Decision Forest (Tong,
2003) method for predicting
estrogen receptor binding activity that also quantifies confidence in
predictions.
back to resources
-
Decision Forest Predictions for 6500 Industrial and
Environmental Chemicals The Decision Forest model with prediction
confidence (Tong,
2004) has been applied to a dataset of 6573
industrial and environmental chemicals. The link below connects to a
separate version of the EDKB database containing only the results of
these predictions.
EDKB Database with Estrogen Activity Prediction of Industrial and
Environmental Chemicals
back to resources
- Four-Phase Screening and Priority Setting Model
Shi (2002), Hong (2002)
and Tong (2002) have described integrated suites of multiple SAR and QSAR models to be used in sequence to prioritize for testing very
large numbers of chemicals based on likelihood of activity. The models
are calibrated to minimize false negative prediction.
back to resources
- Future Web-based prediction The EDKB website plans in the
future to enable website-based models to predict estrogen and androgen
activity based on chemical structure.
back to resources
Please address any questions and suggestions
to Dr. Weida Tong at +1-870-543-7142 (voice) or
weida.tong@fda.hhs.gov.
For any technical problem and reporting
bugs, please send e-mail to
NCTR
Bioinformatics Support .
For more information, please visit the
NCTR/FDA's
Center for Toxicoinformatics.
top
|
