BioLit: integrating biological literature with databases.

BioLit is a web server which provides metadata describing the semantic content of all open access, peer-reviewed articles which describe research from the major life sciences literature archive, PubMed Central. Specifically, these metadata include database identifiers and ontology terms found within the full text of the article. BioLit delivers these metadata in the form of XML-based article files and as a custom web-based article viewer that provides context-specific functionality to the metadata. This resource aims to integrate the traditional scientific publication directly into existing biological databases, thus obviating the need for a user to search in multiple locations for information relating to a specific item of interest, for example published experimental results associated with a particular biological database entry. As an example of a possible use of BioLit, we also present an instance of the Protein Data Bank fully integrated with BioLit data. We expect that the community of life scientists in general will be the primary end-users of the web-based viewer, while biocurators will make use of the metadata-containing XML files and the BioLit database of article data. BioLit is available at http://biolit.ucsd.edu.

INTRODUCTION

Prior to the advent and subsequent popularity of the internet, biological databases and scientific publications were necessarily separate entities. However, now that scientists rely on cyberinfrastructure in the course of their daily research, it is startling that databases and publications are still so distinct from each other. While databases have matured and grown in complexity in the digital age, publications have fared poorly; the traditional concept of an article as a static, independent unit of information persists, albeit in an electronic version. There are myriad ways in which the content of an electronic version of an article can be used in a computational manner and the technology and infrastructure to make this happen already exists; this is demonstrated by, among other things, the existence of biological databases and related data mining tools. Indeed, one hallmark of a successful biological database is integration via data mining with other related databases or resources. Yet integration with literature, which is unquestionably the primary medium through which scientists communicate their research, is conspicuously lacking.

Significant progress in taking advantage of article content is finally being made, perhaps most notably as a result of the open access literature movement pioneered by, among others, the Public Library of Science (PLoS) (1). Articles that are published under an open access license are available for download in toto immediately upon publication and can be distributed freely providing that the original citation is maintained. The deposition of open access life sciences articles has been centralized with the creation of PubMed Central (2) and it is especially important to note that PubMed Central stores many of these articles in a standardized and machine-readable format, which conforms to the National Library of Medicine (NLM) DTD (http://dtd.nlm.nih.gov/). All PubMed Central articles deposited as a result of this movement are thus freely available as XML files, which contain the full-text of the article and some semantic mark-up of the content.

Though a major advantage to both authors and readers, a large segment of the life sciences community remains unaware of the opportunities that are now available due to the existence of PubMed Central as a digital repository, or even of the existence of open access literature altogether. Hopefully, this will change with the recent NIH directive, which mandates the deposition of all findings of NIH-funded research in PubMed Central (http://publicaccess.nih.gov/) and with novel open access-driven applications such as PubNet (3) and SciVee (4). Admittedly, challenges do present themselves when considering the use of open access literature beyond that of a reader or distributor—they have been described elsewhere (4–7)—but opportunities abound and the tenor of current scientific policy and inquiry suggests that the open access corpus will continue to grow significantly (8–10).

BioLit aims to take advantage of the opportunities offered by open access literature by making it possible to include full text or excerpts of these articles directly within existing biological databases and to add newly generated metadata to the articles in order to increase their informative value, tasks which are both made possible by having free access to the full text of the article. These tasks are accomplished by mining the full article text for terms of interest, indexing the terms found, and then including these terms in both machine-readable and human-readable forms that are an enhancement of the original article XML file. The machine-readable form can be incorporated directly into a database or resource. We have prototyped this particular aspect of the effort using a clone of the RCSB Protein Data Bank (PDB) (11), a database of macromolecular structures, and have consequently focused on finding PDB identifiers and Gene Ontology (12) terms as these are both major resources in the structural biology community and the life sciences community as a whole.

Several groups have made significant contributions in applying ontologies to both open- and closed-access biomedical literature. For example, the BioCreAtIvE initiative has spawned a number of tools with the aim of annotating genes and proteins in articles using ontologies and other vocabularies (13). GoPubMed (14) and the more recent SEGOPubmed (15) both aim to add Gene Ontology-based semantic data to PubMed abstracts in order to improve literature searches. The Textpresso team has also focused on improving the classification and searchability of articles by inferring semantic relationships in articles using customized ontologies (16,17). Unlike GoPubMed and SEGOPubmed, they use the full text of the article. Another excellent resource is the National Center for Biomedical Ontology (NCBO) whose overarching aim is to render biomedical information into machine-readable data (18). One of their major contributions is the unification and integration of biomedical ontologies into a single resource, the OBO Foundry, which makes projects such as these and BioLit feasible. The NCBO consortium is also developing a number of web-based tools, in order to make the machine-readable biomedical information available and useful to the community. Finally, The Conceptual Open Hypermedia Service (COHSE), perhaps the most broadly applicable semantic mark-up project, runs as a portlet and adds hyperlinks to existing web pages by allowing a user to select an ontology with terms that can be matched to the page to create links to other pages (19).

Several of the aforementioned tools allow manual curation of the marked-up result in order to increase the accuracy of the mark-up, and hence, the literature search results. This is the ideal approach towards semantic mark-up since the author, or at least an expert human, is providing input. However, the uptake of such an approach in the community has been fairly unsuccessful owing to time limitations on the part of the putative curators and, presumably, lack of demonstrable value to justify the time and effort.

BioLit shares many attributes with these projects but is unique in that it offers a searchable web database of the full text of all PubMed Central research articles with automated mark-up from multiple biomedical ontologies and identifiers from multiple biological databases. No downloading of software is required and all searchable terms belong to established biomedical resources. The focus of this resource is specific to life sciences literature but broad within that realm. Machine-readable files are freely available and distributable via automated means.

BIOLIT DATABASE

The database supporting the BioLit web server contains data from a subset of the PubMed Central holdings. Article archives, including full-text XML files and figures, were retrieved via the PubMed Central FTP site (ftp://ftp.ncbi.nlm.nih.gov/pub/pmc). The articles were first filtered to remove articles that were not labeled as ‘review-article’, ‘case-report’, ‘research-article’, ‘brief-report’, ‘retraction’ and ‘article-commentary’, and documents which contained only scanned content since these are generally advertisements. The remaining articles were stored in a local MySQL database both in the original and parsed formats. When possible, data included in the MEDLINE record for an article were retrieved from PubMed using NCBI's E-Utilities (2). The full text of each article was then parsed to find PDB IDs and Gene ontology terms. These metadata were also stored in the local database. Currently, 51 667 articles exist in the database. The database is updated weekly.

BIOLIT ARTICLE VIEWER

BioLit uses a customized viewer for the articles in order to display the original text inline with the added metadata. This viewer uses a custom XML stylesheet based on the NLM DTD. The full text of the paper can be displayed in a tabbed format or the traditional linear article format and the metadata are highlighted with color and a menu of options when the user clicks on a term. In the case of PDB IDs, an image of the 3D structure identified by the PDB ID is displayed along with links to the macromolecular sequence, the PDB record and other related features. Clicking on an ontology term shows a definition of the term and related links. We also generate statistics describing ontology term usage across all articles and these terms can be used for searching or finding related articles. Figure 1 shows an excerpt of BioLit-enhanced text.

Figure 1.

A screenshot of the BioLit article viewer. This customized viewer not only displays the full text of an article as it was originally published but also displays integrated metadata from the BioLit database. PDB IDs can be seen highlighted in yellow and Gene Ontology terms are highlighted in lavender. Clicking on a highlighted terms brings up information, a list of options or links associated with each term. Article shown is in PLoS Comput. Biol., 2006, 2, e90.

In order to use the BioLit article viewer, the user enters a search term on the BioLit homepage. This term can be an author name, title, keyword, etc. The web server then returns a list of matching articles; clicking on an article invokes the customized viewer. Alternatively, a user could bypass the web-based viewer and retrieve BioLit XML files using a script by appending a PubMed Central ID (PMCID), PubMed ID (PMID) or article digital object identifier (DOI) to the following URL template: http://biolit.ucsd.edu/biolit/getbiolitxml?SOURCE=ID where, SOURCE is ‘pmcid’, ‘pmid’ or ‘doi’ and ID is the corresponding identifier for that resource.

The machine-readable version of an article is available as an XML file through the viewer and via a POST HTTP query. The XML markup enrichment performed by BioLit consists of inserting metadata into the original XML file. The tag appears as the top-level child node of