Final Report
PBPK Literature Visualization Project: ModelMap
Authors:
Susan Ritger Crowell1, Scott Dowson2, and Justin Teeguarden1
1 Systems Toxicology and Exposure Science, Biological Sciences Division, Pacific Northwest
National Laboratory, Richland, WA
2 Visual Analytics, Fundamental and Computational Sciences, Pacific Northwest National
Laboratory, Richland, WA
-------
Background and Objectives
The development and evaluation of PBPK and QSAR models remains a relatively new discipline;
therefore readily accessible and high-quality resources are necessary to support increasing model use.
The resource needs are broad, ranging from parameter values, to model archives and publication
archives. A common thread is the need to organize the available PBPK model related literature in a
fashion that supports rapid access and organization according to data needs and more detailed analysis.
Dr. Rocky Goldsmith, U.S. EPA, presented Battelle with a general plan to create novel
information resources that will support the chemoinformatic needs and or data organization and
archiving in support of the U.S. EPA software program PReParE specifically and PBPK modeling in
general. The series of ideas proposed by Dr. Goldsmith related to the development of an approach to
automate the extraction of data related to PBPK modeling from a body of literature pre-identified as
being related to PBPK modeling (or metabolism, etc.).
Battelle has developed, for other government clients, a number of informatics tools that support
the kind of automated analysis and organization sought by the U.S. EPA for the PBPK modeling
literature. In this work assignment, Battelle modified existing computational and informatics tools to
develop and expand a first generation tool, ModelMap, that can be used to automatically organize and
give rapid access to the body of PBPK modeling literature. The application can be accessed and utilized
at http://pbpk.labworks.org/.
-------
Progress and Workflow
PBPK literature naturally lends itself to hierarchical organization and categorization, features
that we have exploited to create a powerful literature navigation interface. Tree mapping is a method
of information visualization and navigation for hierarchical data such as the PBPK literature: data are
organized thematically and can be navigated by the user choosing a theme of interest, and then drilling
down through sub-themes to the level of individual data or documents.
Hierarchy Development
Literature associated with PBPK and biological modeling was identified using a specific search
string (Appendix) in the PubMed database (http://www.ncbi.nlm.nih.gov/pubmed), yielding an initial
database of > 2,800 MEDLINE abstracts. These data were organized into a logical hierarchy (i.e.,
taxonomy) by subject matter experts (i.e., toxicologists and modelers) using Protege, a publicly available
ontology editor (http://protege.stanford.edu/).
The taxonomy (Figure 1) used to create the tree map reflects the native ways in which users
engage the PBPK related literature. Major categories of interest, called fundamental nodes, were
identified, as well as sub-nodes and leaf nodes, the smallest and most specific unit of the taxonomy. For
example, the fundamental node "Chemicals" contains sub-nodes such as "Volatile Organic Compounds
(VOCs)" and "Pesticides." The sub-node "Pesticides" contains
additional sub-nodes ("Fungicides", "Herbicides", and
"Insecticides"), while the sub-node "VOCs" contains leaf nodes
representing individual compounds (e.g., benzene).
Subsequent to an initial round of taxonomy
development by one subject matter expert, the taxonomy was
reviewed first by an internal panel of subject matter experts at
PNNL, and then by external experts in the field. The internal
panel was comprised of Jordan Smith, Justin Teeguarden, Susan
Crowell, Rick Corley, and Chuck Timchalk (Systems Toxicology
and Exposure Science, PNNL); Dennis McQuerry (Visual
Analytics, PNNL); and Torka Poet (formerly Systems Toxicology
and Exposure Science, PNNL; now Summit Toxicology). The
panel members were briefed on the ModelMap application,
individually reviewed the taxonomy, and then discussed the
taxonomy's structure as a group. The combined comments of
the internal panel were incorporated into the taxonomy, which
was then dispersed to external reviewers. The external
reviewers were Eva McLanahan and Paul Schlosser (US E.P.A),
Jerry Campbell (The Hamner Institute), Hugh Barton (Pfizer),
and Jeff Fisher (NCTR/FDA). Reviewers were again briefed on
the ModelMap application and underlying taxonomy via
webinar, and then invited to submit comments on the
taxonomy. These comments were incorporated into the final
version of the taxonomy delivered to the clients and used in the
ModelMap application.
Literature Search Strings
» •Thing
IT
Elemental!
Endocrinc'Disruptors
Mixtures
NanoparticEes
OrQanornetalMes
• PersistenlOrganicPollurantsPOPs
' Pesticides
Fungicides
Herbicides
DDT
Dieldrin
Qraanophosphates
Chlorpynros
Paratnton
* •Pharmaceuticals
T valatileorganiccernpeunasvOCsl
Benzene
Butadiene
C arbo nTetrach Ig ride
EthaneI
Formaldehyde
Methane I
MethylTerti aryfl utylEth erMTBE
MethyleneChlonde
Perchloreethylene
- TCE
•• Distribution
» Elimination
»• ExposureRoute
Metabolism
»• MedeOrAcoen
» f.lodelSlylt
»• Organism
9 Pftys icalC h em 1C alP rope rt i e 5
» •Platform
Figure 1. Taxonomy of Major PBPK
Themes. Red shaded items were
identified as "Fundamental Nodes" or
themes. Blue shaded items are "Sub-
Nodes" and green shaded items are
"Leaf Nodes."
-------
IN-SPIRE, a unique and powerful data visualization tool that facilitates identification of linguistic
patterns and major themes in data sets that might not be readily apparent (http://in-spire.pnnl.gov/),
was used to analyze abstracts for key terms for further development and refinement of the taxonomy,
as well as to develop and test search strings to isolate specific documents relevant to each leaf node of
the hierarchy (Figure 2). For example, the search string used to identify abstracts under the leaf node
"Sprague Dawley Rat" (Taxonomy path: Organism > Rodent > Rat > Sprague Dawley) is:
("SD (rat rats)" "sprague dawley" sprague-dawley)
Entering this search string into IN-SPIRE isolates the 196 abstracts focused on PBPK models or relevant
research conducted using Sprague Dawley rats (Figure 3). Developers and subject matter experts
worked together with IN-SPIRE to generate appropriate search strings for each of the leaf nodes
identified in the taxonomy. This iterative process involved testing proposed search strings against the
database of MEDLINE abstracts, assessing the returned documents for relevance, refining the search
strings based on these assessments, and so forth.
Once developed, search strings were incorporated directly into the taxonomy within Protege,
the ontology editor.
-------
Figure 2A. IN-SPIRE galaxy view of PBPK literature space, with documents relevant to Sprague-Dawley rats
highlighted in green.
°r?-.'*W
• l(KmW* QWt- M»
I* 106600.11 AphrmA;i)KJ»r'ijndp*i*
. IMC-Wi] CtlnnO I-. IMO-4M1
^jwn«_r»,,!W7)i?i
..»onyi.'i«ll*
.trt_A*r.-'9MMJ.
Figure 2B. IN-SPIRE document viewer, focused on documents relevant to Sprague-Dawley rats.
-------
Tree Map Visualization
The publications from PubMed are obtained through email in a text format:
PMID- 24671884
OWN-NLM
STAT- Publisher
DA - 20140327
This data is converted to XML and goes through a tagging process based on the search strings defined
for each leaf node in the taxonomy. A utility that is a part of IN-SPIRE, called Taggert, processes an XML
document, and analyzes it in the context of a supplied taxonomy. The text of each publication is
extracted and tested against each node's search string. When a match is detected, for example:
Node: (Organism > Rodent > Rat > Sprague Dawley)
Search String: ("SD (rat rats)" "sprague dawley" sprague-dawley)
Abstract:
... PET radiotracerfor mGlu4, and characterized its biological properties in Sprague Dawley rats.
[(11)C]3 was synthesized from ...
That abstract will be annotated with the path (Organism/Rodent/Rat/Sprague Dawley). As the tagging
process is solely driven by the structure and search strings in the taxonomy, a publication may by tagged
with multiple taxonomy paths, or none. Based on all the tags captured in the set of publications, a tree
map visualization is then used to organize and display the set of abstracts in the context of the
organizational structure of the taxonomy.
The finished ModelMap application can be found at http://pbpk.labworks.org/. The tree map style of
visualization shows the distribution of publications based on their associated tags at a specified node in
the taxonomy, and allows the user to navigate rapidly through subjects of interest to the associated
abstracts. Initially the application opens to shows the distribution at the fundamental node, i.e., the
highest level of categorization. Each colored [fundamental node] box (e.g. 'Chemicals', 'Model Style',
'Exposure Route') is sized so that the area is proportional to the number of abstracts tagged as relevant
to that category. The tree map visualization also shows the sub-nodes of each fundamental node, using
various shades of the fundamental node's color (Figure 4.). This provides more insight into the
distribution of the data beyond the parent node. The visualization supports navigation and drill-down
through the taxonomy by allowing users to click on a particular [fundamental] node which resets the
visualization anchoring on the selected node, and then showing the distribution of data on the sub-node
and leaf node level.
-------
Figure 4. Tree map of PBPK Literature. Each labeled box corresponds to a major theme. Clicking within
a box of interest allows the user to drill down to greater levels of detail, eventually to the level of
relevant individual documents.
Autoupdate Features
The search string used to identify the initial literature database was saved in PubMed using a
dedicated email address (pbpkharvest@pnnl.gov). Every day, PubMed emails a text report of all new
publications that match the saved search. An automated process downloads these emails, parses the
reported abstracts into an XML format, tags them according to the existing ModelMap ontology, and
finally updates the final collection of all archived publications.
Additional Documentation
Installation details can be found within the delivered code, and at
http://srs.pnnl.gOV/wiki/index.php/Admin:installing-SRS.
-------
Appendix
Search string used to identify relevant literature in PubMed:
pbpk[AII Fields] OR pbtk[AII Fields] OR ((physiologic[Text Word] OR physiologically[Text Word]
OR physiological [Text Word]) AND based[AII Fields] AND (pharmacokinetic[Text Word] OR
toxicokinetic[Text Word] OR pharmacokinetics[Text Word]) AND (model[Text Word] OR
models[Text Word] OR modelling[Text Word])) AND English[lang]
------- |