An open source infrastructure for managing knowledge and finding potential collaborators in a domain-specific subset of PubMed, with an example from human genome epidemiology
© Yu et al; licensee BioMed Central Ltd. 2007
Received: 10 April 2007
Accepted: 09 November 2007
Published: 09 November 2007
Identifying relevant research in an ever-growing body of published literature is becoming increasingly difficult. Establishing domain-specific knowledge bases may be a more effective and efficient way to manage and query information within specific biomedical fields. Adopting controlled vocabulary is a critical step toward data integration and interoperability in any information system. We present an open source infrastructure that provides a powerful capacity for managing and mining data within a domain-specific knowledge base. As a practical application of our infrastructure, we presented two applications – Literature Finder and Investigator Browser – as well as a tool set for automating the data curating process for the human genome published literature database. The design of this infrastructure makes the system potentially extensible to other data sources.
Information retrieval and usability tests demonstrated that the system had high rates of recall and precision, 90% and 93% respectively. The system was easy to learn, easy to use, reasonably speedy and effective.
The open source system infrastructure presented in this paper provides a novel approach to managing and querying information and knowledge from domain-specific PubMed data. Using the controlled vocabulary UMLS enhanced data integration and interoperability and the extensibility of the system. In addition, by using MVC-based design and Java as a platform-independent programming language, this system provides a potential infrastructure for any domain-specific knowledge base in the biomedical field.
Published literature databases are a major information source for generating scientific hypotheses and conducting evidence-based reviews . PubMed/Medline is the largest published literature repositories in the biomedical world, containing more than 15 million citations from about 5000 journals . The ever-increasing quantity of published literature creates challenges for searching and data-mining. Many efforts have attempted to improve the search capacity and performance of PubMed, for example, by creating an alternative, user-friendly Web interface , adopting a semantics-based ranking algorithm , or using a BLAST-style text similarity search algorithm . Several biomedical research fields have also generated domain-specific, Web-based information resources by collecting and curating PubMed citations and other data relevant to their interests [6–8]. These smaller, more specific information sources can be more easily queried and used for domain-specific data mining.
The human genome epidemiology literature database  is a domain-specific, published-literature database created originally in 2001, sponsored by the National Office of Public Health Genomics at the Centers for Disease Control and Prevention. By collecting and curating citations from PubMed that report epidemiologic analyses of gene-disease associations, the database facilitates meta-analyses in the rapidly emerging field of human genome epidemiology (HuGE). The database currently contains more than 30,000 citations with more than 6,000 new articles added each year .
We present an open source infrastructure for managing and querying domain-specific data (Literature Finder) and investigator information (Investigator Browser) available in PubMed abstracts, along with a data management and curating tool set. We illustrate these functions using a knowledge base system called HuGE Navigator  which was built upon this open source infrastructure. By integrating the Unified Medical Language System (UMLS) into our open source package, we provide a novel IT infrastructure that facilitates data integration, interoperability, and allows for future expansion to include additional applications.
The Web-based infrastructure for this system was designed to manage a local collection of PubMed literature and generate investigator profiles based on authorship and affiliation information. The main objectives of the design were data standardization, automatic capacity for data manipulation, modularity and scalability of the system, and a user-friendly Web interface.
Data management implementation
The UMLS  contains more than 100 vocabularies from biomedical fields; the many synonyms and variants for each unique concept are linked with a UMLS concept unique identifier (CUI). Indexing the literature using UMLS CUIs enhances interoperability and integration of the data, while increasing the sensitivity of data retrieval and allowing for robust free text searching and system extensibility.
To increase the granularity of relationships among the millions of unique concepts collected in UMLS, we used the Medical Subject Headings (MeSH) hierarchy tree to establish parent-child relationships. MeSH  indexed by PubMed staff are converted to UMLS CUIs automatically, reducing the manpower needed in the curating process.
We enriched the gene information in the UMLS metathesaurus by incorporating data downloaded from Entrez Gene records , substituting Entrez Gene IDs for the UMLS CUIs. Although HUGO, a nomenclature for human genes , is one of the controlled vocabularies in the UMLS, we found that Entrez Gene was more comprehensive than HUGO, including more gene aliases and additional genetic information, such as chromosome location and OMIM ID.
To improve the performance of the system, we designed a dynamic data subset-creating process for the external datasets. The 6 million records in the UMLS concept-lookup table could create performance issues if queried directly. Even after removing non-English and retired terms, the table contained 3 million records. Because of the multidisciplinary nature of HuGE research, we could not further subset the UMLS metathesaurus by semantic type. Since not all 3 million UMLS terms will be used in literature indexing, we developed a UMLS concept subset-creating script as part of the curating utilities. This script populates the subset table automatically and dynamically by adding any newly encountered terms into the UMLS subset table in the database when PubMed records are uploaded or updated (see Curating utility implementation section). Applying this process dramatically reduced the size of the UMLS subset table to about 23,000 records, significantly improving performance. The same mechanism was applied to the MeSH hierarchy data, creating subsets that are used to retrieve children terms for the query.
Application feature implementation
Main literature search features
The simple search using the Search Text box returns all available articles that meet the query specification, which can be displayed with 25, 50, or 200 records on a page. The PubMed It feature opens a PubMed page with the returned articles. The user can then search with features provided by PubMed or upload the list of PubMed abstracts into reference software such as Endnote and Reference Manager (Figure 1).
Seven classifiers (Disease, Exposure, Gene, Study Type, Category, Year, Author) are available for the filtering feature in the HuGE Literature Finder (Figure 1).
Main investigator search features
A list of investigators with their corresponding number of publications is returned based on the user's query in the text search box (Figure 2). The user can choose to display all authors or first and last authors only. The list can be sorted alphabetically by author's name or by the number of publications.
The filtering feature in the Investigator Browser includes two classifiers: Country and Institute (Figure 2).
Curating utility implementation
A number of tools were developed to curate the database automatically. These consist of the following:
The PubMed literature loader automatically uploads records from the PubMed database into the local database using NCBI E-Utilities  based on PubMed IDs. Data including the title, abstract, author (first initial, last name), journal (name, volume, issue), publication date (month, year) and affiliation string are used to populate the corresponding database tables.
The MeSH index loader automatically uploads MeSH terms provided in the PubMed record into the database when these terms are available.
The MeSH-UMLS converter converts and maps MeSH terms to the corresponding UMLS CUI.
The UMLS/Entrez Gene subset-generator automatically creates a UMLS/Entrez Gene table with subsets based on terms used in the database.
The MeSH Tree subset-generator automatically creates MeSH Tree table with subsets based on the terms used in the database.
The affiliation parser automatically parses the author affiliation string into full mail address, institution, country and email address, and populates the database with the parsed information. The detailed methodology has been reported .
The whole infrastructure can be divided into three discrete modules that are loosely coupled. The data module contains all data in the database; the accessory utility module is responsible for a series of data transactions and manipulations; and the application module includes all the applications in the system. To avoid version control problems, we allow data entities from external data sources (e.g., UMLS Metathesaurus, Entrez Gene, and MeSH Tree) to be updated as needed without an overhaul of the entire system. Each application was built on this model, allowing for seamless navigation and easy plug-in of new applications.
Database schema implementation
Information retrieval preliminary evaluation
To test the system's information retrieval performance, we first populated it with 500 randomly selected articles from the human genome epidemiology literature database (HuGE Literature Finder). Independently, two of us (W.Y., A.Y.) assessed all 500 abstracts for relevance to any of the five diseases or the five genes that appear with greatest frequency in the database . All discrepancies were discussed and a final consensus was reached for each article. We then queried the system using the same 10 terms and compared the results with our independent assessment. By using the method described by Zhou, et al. , we estimated system recall to be 90% and precision 93%. The formulas to calculate are as followed:
where TP, FP and FN represent the number of true positives, false positives and false negatives respectively.
Usability test results.
The interface of this system is pleasant
I can perform complex searches
It is easy to find the information I needed
I feel comfortable using this system
I can effectively obtain information
System speed is reasonable
Easy to learn how to use it
Overall, I am satisfied with how easy it is to use this system
With advances in Web technology, online searching has become one of the most preferred methods for obtaining information in the health science setting . Although the PubMed/Medline database provides users with a central place to search the biomedical literature, efficient and effective direct searches, using simple key words or complex queries, are often challenging. We have created an open source infrastructure to manage and build a Web-based, domain-specific database from PubMed records. Integrating information from UMLS and Entrez Gene enhances both the sensitivity of the HuGE Literature Finder and the information available to the user. The application infrastructure also provides data mining capacity that automatically extracts investigator profile including mailing address, institution, country and email address from the authorship and affiliation information provided in PubMed abstracts. Recently, developing investigator collaborative networks has become an important agenda in the field of human genome epidemiology, to promote collaborations, facilitate the standardization of study design and analytical methods, confirm findings, and produce systematic reviews [10, 25].
There have been many initiatives in the goal of improving the PubMed data retrieval, such as SLIM  that enhances the usability of the PubMed, and PubFocus that prioritized the Medline/PubMed record based on the statistical analysis of the query and other factors . A most recent activity is the Semantic Medline developed by the National Library of Medicine that uses the natural language processing technique to predict the semantic relationship of the query to other biologic concepts . As contrast with those initiatives that intend to work on the whole PubMed/Medine, this open source infrastructure is focusing on managing the highly curated PubMed data for a specific biomedical domain and make source codes available to the biomedical community to build their own specialized web-based database easily, simple and intuitive user interfaces increase the usability of information systems. The design of our Web interface accommodates the tendency of most users to search published literature by simple keywords, then filtering down through the retrieved results. The usability test demonstrated that most appreciated this aspect of the interface design.
Modularity and scalability in the MCV-based design of the infrastructure will allow the system to expand easily as needed. Any individual application with specific business logic and requirements can be plugged into the system. We have experimented with this idea by adding supplementary applications, such as other components of HuGE Navigator . A critical feature of this infrastructure application is the use of a robust controlled vocabulary to standardize the data. Because PubMed, Entrez Gene, and UMLS are integrated into the indexing mechanism, the infrastructure of this system is extensible beyond the literature indexing provided by MeSH. For example, integrating laboratory information management into the system would be simple because SNOMED , one of the controlled vocabulary collections in UMLS, is suitable for laboratory data. UMLS has successfully demonstrated the ability to map many different controlled vocabularies into the standard vocabulary provided by UMLS concepts [28, 29]. The design aims to achieve full integration and interoperability at both the system and data levels.
The open source system infrastructure presented in this paper provides a novel approach to managing and querying information and knowledge from domain-specific PubMed data. To ensure data integration, interoperability, and system extensibility, we have developed a novel approach to generate dynamically a controlled vocabulary for a specific biomedical domain. A performance evaluation of the system demonstrated high recall and precision. Results of usability testing showed that the Web interface was easy to learn and queries to be completed queries quickly and effectively. The ability to generate a dynamic controlled vocabulary, the MVC-based design, and Java as a platform-independent programming language allow this infrastructure to be used for other domain-specific knowledge bases in the biomedical field.
Availability and requirements
The system built upon this open source infrastructure: http://www.hugenavigator.net
Operating systems: Windows and Linux/Unix
Database: MS SQL server and MySQL
Programming language: Java
Software packages: J2EE 1.4, Hibernate 3.0 and Strut 1.2.9
License: GNU General Public License. This license allows the source code to be redistributed and/or modified under the terms of the GNU General Public License as published by the Free Software Foundation. The source code for the application is available at no charge.
Any restrictions to use by non-academics: None
Concept Unique Identifier
Human Genome Epidemiology
The Human Genome Organization
Medical Subject Heading
Online Mendelian Inheritance in Man
Unified Medical Language System
Disclaimer: The findings and conclusions in this report are those of the author(s) and do not necessarily represent the views of CDC
- Potter B, Rotert E: Making evidence-based practice a reality. WMJ 2005, 104: 22–24.PubMedGoogle Scholar
- [http://www.ncbi.nlm.nih.gov/entrez]PubMed. Bethesda, MD: National Library of Medicine; 2006.http://www.ncbi.nlm.nih.gov/entrez
- Muin M, Fontelo P, Liu F, Ackerman M: SLIM: an alternative Web interface for MEDLINE/PubMed searches – a preliminary study. BMC Med Inform Decis Mak 2005, 5: 37. 10.1186/1472-6947-5-37PubMed CentralView ArticlePubMedGoogle Scholar
- Plikus MV, Zhang Z, Chuong CM: PubFocus: semantic MEDLINE/PubMed citations analytics through integration of controlled biomedical dictionaries and ranking algorithm. BMC Bioinformatics 2006, 7: 424. 10.1186/1471-2105-7-424PubMed CentralView ArticlePubMedGoogle Scholar
- Lewis J, Ossowski S, Hicks J, Errami M, Garner HR: Text similarity: an alternative way to search MEDLINE. Bioinformatics 2006, 22: 2298–2304. 10.1093/bioinformatics/btl388View ArticlePubMedGoogle Scholar
- Human Genome Epidemiology NetworkAtlanta, GA: National Office of Public Health Genomics, Centers for Disease Control and Prevention; [http://www.cdc.gov/genomics/hugenet/default.htm]
- Bertram L, McQueen M, Mullin K, Blacker D, Tanzi R: The AlzGene Database. Alzheimer Research Forum [http://www.alzgene.org]
- The UCSD-Nature Signaling Gateway[http://www.signaling-gateway.org/]
- Lin BK, Clyne M, Walsh M, Gomez O, Yu W, Gwinn M, Khoury MJ: Tracking the epidemiology of human genes in the literature: the HuGE Published Literature database. Am J Epidemiol 2006, 164: 1–4. 10.1093/aje/kwj175View ArticlePubMedGoogle Scholar
- Ioannidis JP, Gwinn M, Little J, Higgins JP, Bernstein JL, Boffetta P, Bondy M, Bray MS, Brenchley PE, Buffler PA, Casas JP, Chokkalingam A, Danesh J, Smith GD, Dolan S, Duncan R, Gruis NA, Hartge P, Hashibe M, Hunter DJ, Jarvelin MR, Malmer B, Maraganore DM, Newton-Bishop JA, O'Brien TR, Petersen G, Riboli E, Salanti G, Seminara D, Smeeth L, Taioli E, Timpson N, Uitterlinden AG, Vineis P, Wareham N, Winn DM, Zimmern R, Khoury MJ, Human Genome Epidemiology Network and the Network of Investigator Networks: A road map for efficient and reliable human genome epidemiology. Nat Genet 2006, 38: 3–5. 10.1038/ng0106-3View ArticlePubMedGoogle Scholar
- HuGE Literature Finder[http://www.hugenavigator.net/HuGENavigator/startPagePubLit.do]
- HuGE Navigator[http://www.hugenavigator.net/]
- Lindberg DA, Humphreys BL, McCray AT: The Unified Medical Language System. Methods Inf Med 1993, 32: 281–291.PubMedGoogle Scholar
- Medical Subject Heading[http://www.nlm.nih.gov/mesh/]
- Entrez Gene[http://www.ncbi.nlm.nih.gov/entrez/query.fcgi?DB=gene]
- HUGO Gene Nomenclature CommitteeLondon, United Kingdom: The Human Genome Organisation; 2004. [http://www.hugo-international.org/committee_nomen.htm]
- NCBI E-utilities. ESpell[http://eutils.ncbi.nlm.nih.gov/entrez/query/static/espell_help.html]
- Entrez Programming Utilitiesbethesda, MD: National Library of Medicine; 2006. [http://eutils.ncbi.nlm.nih.gov/entrez/query/static/eutils_help.html]
- Yu W, Yesupriya A, Wulf A, Qu J, Gwinn M, Khoury MJ: An automatic method to generate domain-specific investigator networks using PubMed abstracts. BMC Med Inform Decis Mak 7(1):17. 2007 Jun 20; 2007 Jun 20; 10.1186/1472-6947-7-17
- Singh I, Stearns B, Johnson M, Enterprise Team: Designing Enterprise Applications with the J2EE Platform. Addison-Wesley Publishing Co., Reading, MA, 2002; 2006. ISBN: 0201787903. ISBN: 0201787903.Google Scholar
- Tai H, Mitsui K, Nerome T, Abe M, Ono K, Hori M: Model-driven development of large-scale Web applications. IBM Journal of Research and Development 2004.Google Scholar
- Zhou W, Smalheiser NR, Yu C: A tutorial on information retrieval: basic terms and concepts. J Biomed Discov Collab 2006, 1:2: 2. 10.1186/1747-5333-1-2View ArticleGoogle Scholar
- Jamieson S: Likert scales: how to (ab)use them. Med Educ 2004, 38: 1217–1218. 10.1111/j.1365-2929.2004.02012.xView ArticlePubMedGoogle Scholar
- De Groote SL, Dorsch JL: Measuring use patterns of online journals and databases. J Med Libr Assoc 2003, 91: 231–240.PubMed CentralPubMedGoogle Scholar
- Ioannidis JP, Gwinn M, Little J, Higgins JP, Bernstein JL, Boffetta P, Bondy M, Bray MS, Brenchley PE, Buffler PA, Casas JP, Chokkalingam A, Danesh J, Smith GD, Dolan S, Duncan R, Gruis NA, Hartge P, Hashibe M, Hunter DJ, Jarvelin MR, Malmer B, Maraganore DM, Newton-Bishop JA, O'Brien TR, Petersen G, Riboli E, Salanti G, Seminara D, Smeeth L, Taioli E, Timpson N, Uitterlinden AG, Vineis P, Wareham N, Winn DM, Zimmern R, Khoury MJ, Human Genome Epidemiology Network and the Network of Investigator Networks: A network of investigator networks in human genome epidemiology. Am J Epidemiol 2005, 162: 302–304. 10.1093/aje/kwi201View ArticlePubMedGoogle Scholar
- Semantic MEDLINE[http://skr3.nlm.nih.gov/SemMedDemo/]
- Sarkar IN, Cantor MN, Gelman R, Hartel F, Lussier YA: Linking biomedical language information and knowledge resources: GO and UMLS. Pac Symp Biocomput 2003, 439–450.Google Scholar
- Ingenerf J, Reiner J, Seik B: Standardized terminological services enabling semantic interoperability between distributed and heterogeneous systems. Int J Med Inform 2001, 64: 223–240. 10.1016/S1386-5056(01)00211-8View ArticlePubMedGoogle Scholar
This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.