FACTA: a text search engine for finding associated biomedical concepts

Yoshimasa Tsuruoka 1 2 Jun'ichi Tsujii 0 1 2 Sophia Ananiadou 1 2 Associate Editor: Jonathan Wren 0 Department of Computer Science, The University of Tokyo , Japan 1 National Centre for Text Mining (NaCTeM) , Manchester, UK 2 School of Computer Science, The University of Manchester Summary: FACTA is a text search engine for MEDLINE abstracts, which is designed particularly to help users browse biomedical concepts (e.g. genes/proteins, diseases, enzymes and chemical compounds) appearing in the documents retrieved by the query. The concepts are presented to the user in a tabular format and ranked based on the co-occurrence statistics. Unlike existing systems that provide similar functionality, FACTA pre-indexes not only the words but also the concepts mentioned in the documents, which enables the user to issue a flexible query (e.g. free keywords or Boolean combinations of keywords/concepts) and receive the results immediately even when the number of the documents that match the query is very large. The user can also view snippets from MEDLINE to get textual evidence of associations between the query terms and the concepts. The concept IDs and their names/synonyms for building the indexes were collected from several biomedical databases and thesauri, such as UniProt, BioThesaurus, UMLS, KEGG and DrugBank. Availability: The system is available at http://www.nactem.ac.uk/ software/facta/ Contact: 1 INTRODUCTION Information about pairwise association between biomedical concepts, such as genes, proteins, diseases and chemical compounds constitutes an important part of biomedical knowledge.1 It is common for a researcher to need answers to questions like What diseases are relevant to a particular gene? or What chemical compounds are relevant to a particular disease? Text mining complements biomedical databases by providing researchers with a convenient way to find such information from the literature. There are a number of web-based text mining applications which can be used for this purpose. EBIMed (RebholzSchuhmann et al., 2007) receives a PubMed-style query from To whom correspondence should be addressed. 1In this article, a biomedical concept refers to a conceptual entity which is normally grounded to a record in a biomedical database. In text, the same concept (e.g. UniProt:O00203) may be represented by different terms (e.g. AP-3 complex subunit beta-1 or Beta3A-adaptin). Note also that the same term may represent different concepts depending on the context, although this problem is currently not resolved in FACTA. the user and analyzes the matched documents to recognize protein/gene names, GO annotations, drugs and species mentioned. Frequently occurring concepts are shown in a table, and the user can view the sentences corresponding to the associations. PolySearch (Cheng et al., 2008) can produce a list of concepts which are relevant to the users query by analyzing multiple information sources including PubMed, OMIM, DrugBank and Swiss-Prot. It covers many types of biomedical concepts including diseases, genes/proteins, drugs, metabolites, SNPs, pathways and tissues. Systems that provide similar functionality include XplorMed (Perez-Iratxeta et al., 2003), MedlineR (Lin et al., 2004), LitMiner (Maier et al., 2005) and Anii (Jelier et al., 2008). Although these applications are useful in exploring such information in the literature, not many of them provide real-time responsesthe users often have to wait for several minutes (or even hours) before they receive the results. Some of the systems provide reasonably quick responses by limiting the number of documents to be analyzed to a very small number (e.g. 500 abstracts), but such limitation leads to a significant deterioration of the coverage. LitMiner and Anii are exceptions in that they can return the result immediately, presumably thanks to pre-computed association statistics between the concepts. However, they do not accept a flexible query (e.g. free keywords or Boolean combinations of keywords/concepts), hence the concepts that can be specified by the users query are limited to predefined ones. To complement existing applications, we have developed FACTA, which is a text search engine for browsing biomedical concepts that are potentially relevant to a query. The distinct advantage of FACTA is that it delivers real-time responses while being able to accept flexible queries. This is achieved by online computation of association statisticsFACTA analyzes the documents retrieved by the query dynamically, using pre-indexed words and concepts. SOFTWARE FEATURES FACTA receives a query from the user as the input. A query can be a word (e.g. p53), a concept ID (e.g. UNIPROT:P04637), or a combination of these [e.g. (UNIPROT:P04637 AND (lung OR gastric))]. The system then retrieves all the documents that match the query from MEDLINE using word/concept indexes. The concepts contained in the documents are then counted and ranked Fig. 1. A screenshot of FACTA search results. according to their relevance to the query. The results are presented to the user in a tabular format. Figure 1 shows an example of the search result. For the input query apoptosis AND blood, the system retrieved 7734 documents from MEDLINE in 0.04 s. The relevant concepts of six categories are displayed in a table and ranked by their frequencies. The document icon next to each concept name in the table allows the user to view snippets from MEDLINE and see textual evidence of the association. The user can also invoke another search by clicking a concept name in the table. This allows the user to explore associations between many different concepts in a highly interactive manner. FACTAs real-time responses to the queries are made possible by the use of its own indexing scheme and implementation of the analysis engines in C++. It uses two indexes built offlineone for the words and the other for the concepts. Both indexes are stored in memory to achieve quick responses, while the actual sentences of MEDLINE abstracts are stored on external storage. The system runs on a generic Linux server with 2.2 GHz AMD Opteron processors and 16 GB memory. Currently, FACTA covers six categories of biomedical concepts: human genes/proteins, diseases, symptoms, drugs, enzymes and chemical compounds. The concepts appearing in the documents are recognized by dictionary matching. In total, 80 260 unique concepts are indexed. We used UniProt accession numbers as the concept IDs for genes/proteins and collected their names and synonyms from BioThesaurus (Liu et al., 2006). We used UMLS (Humphreys and Lindberg, 1989) for diseases and symptoms. The concept IDs and names for drugs, enzymes and chemical compounds were collected from several databases including HMDB, KEGG and DrugBank. Ambiguity causes problems in indexing. For example, the term collapse is not necessarily used as a symptom name in the documents that produced the results shown in Figure 1, so ideally such occ (...truncated)