LigerCat is a web application that allows users to search the PubMed database for articles using keywords or DNA/protein sequences. It displays the search results as a tag cloud of associated MeSH descriptors weighted by frequency, and as a publication timeline graph. Users can explore subsets of articles by selecting terms from the cloud. LigerCat integrates local databases with NLM resources to provide precompiled queries and dynamically update results. It uses a scalable architecture with load balancing, queues, and Redis to improve processing speeds.

1. LigerCat
Holly Miller, MBLWHOI Library, Marine Biological Laboratory
Catherine N. Norton, MBLWHOI Library, Marine Biological Laboratory
Indra Neil Sarkar, Biomedical Informatics, University of Vermont
LigerCat (http://ligercat.org) is a web application that allows a user to search for
articles in PubMed with key words or even a DNA/protein sequence and then
explore the resulting set via the MeSH descriptors and a publication time line.
LigerCat is freely available on the Internet and the source code is available under an
MIT license (http://github.com/mbl-cli/LigerCat).
LigerCat, Literature and Genomic Electronic Resource Catalogue, aggregates
multiple articles in PubMed, combining the associated MeSH descriptors into a tag
cloud, weighted by frequency -- more frequent terms are in a larger font size.
Through LigerCat a user can search the PubMed database using plain language text
or molecular sequences. The MeSH tag cloud then allows the user to explore the
collection and create subsets of articles simply by clicking on terms in the cloud.
The three ways to search in LigerCat: Journals, Articles and Genes.
Journal Search
Queries entered into the LigerCat “Journals” tab are used to search the NLM Journals
database. Searches are enhanced using words in titles of journals. This approach has
shown promise to categorize biomedical literature. In our implementation, stop
words (e.g., “the,” “journal,” “association”) are removed from the list of journal titles
returned from the initial search, and then the top 15% of remaining title words are
determined. The initial search is then enhanced with these additional title words.
Across all the journals indexed in MEDLINE, we observed that on average this
enhanced approach increased results by approximately 25% over just using subject
terms associated with journals. In a specific example, searching the NLM Journals
database for journals using the term “Aging” returns 41 journals; the enhanced
approach implemented in LigerCat returns 130.
LigerCat presents users with the resulting list of journal titles, and the user selects
those that they wish to include in the MeSH cloud generation process. Using a local
MEDLINE database, all PubMed identifiers (PMIDs) are then retrieved for the
selected list of journals.
The MeSH cloud is interactive. The user can click on terms in the cloud to create
subsets of articles in real time. The resulting subset can be displayed in PubMed by
clicking on “Go to Pubmed.”
Article Search
Traditional PubMed searches are fundamental to the identification of relevant
biomedical knowledge. Through the “Articles” tab in LigerCat, users enter a query as

2. they would enter in the PubMed interface. A series of eUtils-mediated searches are
then used to retrieve the list of PMIDs. In addition to the MeSH cloud, a Publication
History graph is displayed. Both features are interactive. Moussing over the
histogram will highlight years and display the number of articles from that year that
your search retrieved. Clicking on the bar for that year performs the search in
PubMed and takes you to that result page.
Gene Search
The third modality of search aims to identify relevant literature that is associated
with a molecular sequence. Within the context of GenBank, 71% of sequences with
citation information are associated with at least one MEDLINE citation. Each PMID
reference associated with GenBank records are extracted and stored within a local
GenBank database. Through the LigerCat “Genes” tab, the user can enter a FASTA-
formatted sequence of a single molecule (e.g., a gene). Related sequences are then
identified using netBLAST. Based on the returned results, sequences are examined
above a minimally stringent E-value (1.0). A list of PMIDs is then derived from the
local GenBank database. The MeSH descriptors from those articles are displayed as a
MeSH cloud similar the one generated from the Article Search. In addition to the
MeSH cloud, a Publication History graph is displayed, however, at this point it is not
interactive.
Under the hood
LigerCat makes use of a hybrid approach of local databases and Web accessible
resources at the National Library of Medicine (NLM Journals, PubMed, and
GenBank). This synergy between live and archived data is leveraged both for speed
(many of the queries and MeSH profile data are precompiled) and to maintain
currency. Additionally, the use of the eUtils for mediating MEDLINE queries enables
the leveraging of the synonym mapping features that are part of the Entrez
interface. As new items are discovered through the system, they are dynamically
added to the database and the MeSH ranks are appropriately adjusted.
LigerCat uses a fast and scalable architecture. The queries to NLM are mediated
through a load-balancing mechanism that off-loads tasks to a background queue.
This is necessary in part because eUtils requires that there is a minimum 1-second
interval between queries to the live data and the volume of data that may need to be
processed to create the resultant MeSH cloud. LigerCat utilizes a number of open
source tools to process data in parallel. By designing a modular, queue-based
architecture, we were able to eliminate single points of failure in the system while
also gaining the benefit of horizontal scalability. The modular nature of the
architecture allowed us to replace a traditional relational database system with the
Redis in-memory key-value store affording us a five fold decrease in processing
times, though at the cost of an increased memory requirement. By modularizing and
virtualizing components of the architecture, we were able to scale processing across
many processor cores of a single or many hosts or out onto public clouds where
required. Scaling can occur in parallel and rapidly due to the small footprint of the
processing components. LigerCat was developed using the Ruby on Rails

3. framework, which allowed for rapid prototyping, development, testing and
integration of the system’s various components. The use of messaging queues, in-
memory key-value stores and related tools offers informatics developers new
solutions to existing data management and processing challenges. The modular
nature enables components of this existing system to be reused in other research
projects that are faced with similar challenges.
Acknowledgements
LigerCat development was made possible through funds from The Ellison Medical
Foundation, NIH, and the Marine Biological Laboratory.