A Statistical Approach to Scanning the Biomedical Literature for Pharmacogenetics Knowledge

Design: The authors built and evaluated several candidate statistical models that characterize pharmacogenetics articles in terms of word usage and the profile of Medical Subject Headings (MeSH) used in those articles. The bestperforming model was used to scan the entire Medline article database (11 million articles) to identify candidate pharmacogenetics articles. Results: A sampling of the articles identified from scanning Medline was reviewed by a pharmacologist to assess the precision of the method. The authors’ approach identified 4,892 pharmacogenetics articles in the literature with 92% precision. Their automated method took a fraction of the time to acquire these articles compared with the time expected to be taken to accumulate them manually. The authors have built a Web resource (http://pharmdemo.stanford. edu/pharmdb/main.spy) to provide access to their results. Conclusion: A statistical classification approach can screen the primary literature to pharmacogenetics articles with high precision. Such methods may assist curators in acquiring pertinent literature in building biomedical databases. j J Am Med Inform Assoc. 2005;12:121–129. DOI 10.1197/jamia.M1640. A challenge for biomedical researchers in the postgenomic era is to use prior knowledge about pharmacogenetics to understand the genetic basis for drug response and to predict drug response in individual patients. Pharmacogenetics research studies investigate how variations in particular genes alter the efficacy and toxicity of drugs. These studies contribute knowledge about which genes are involved in producing or altering a drug effect (mechanism of action or metabolism) and how genetic variations alter observable responses. There are more than 3million citations inMedline mentioning the term gene or drug, and a resource that summarizes the gene–drug relationships that have been established and the level of support for them in the literature would help researchers identify gaps in the current knowledge and plan new research studies. Other text sources such as drug patent literature or drug information databases could also provide useful data for this resource. Many databases collect genetic and protein sequences, structures, expression, and phenotype data, but few contain information that would permit scientists to explore possible connections between genes and drugs. Pharmacogenetics data are just beginning to become available online, and most are in unstructured free-text format. Those few resources that contain pharmacogenetically relevant data focus on a subset of pharmacogenetics studies. A comprehensive resource of information that relates all genes and drugs does not yet exist. We have been building PharmGKB, a resource for pharmacogenetics. One of the services PharmGKB provides is a curated repository of gene–drug relationships, annotated with literature support, contributed by curators and the scientific community. While growth of our repository has been increasing progressively, the ability to create a comprehensive resource is limited by the enormity of the literature and the time it takes curators to identify relevant articles. The medical literature has been the primary venue for distributing the results of pharmacogenetics studies. One possibility for building a pharmacogenetics resource is to populate it directly from the literature by screening all published articles for pertinent pharmacogenetics content. Some biomedical Affiliations of the authors: Section of Medical Informatics, Stanford University, Stanford, CA (DLR, TEK, RBA); Department of Genetics, Stanford Medical Center, Stanford, CA (CFT, TEK, RBA). This work is supported by grants from the National Institute of General Medical Sciences (NIGMS), Human Genome Research Institute (NHGRI), National Library of Medicine (NLM), and the NIH/NIGMS Pharmacogenetics Research Network and Database (U01GM61374). The authors thank John Conroy for his expertise and assistance in accessing the PharmGKB database. Correspondence and reprints: Daniel L. Rubin, MD, MS, Section of Medical Informatics, MSOB X-215, Stanford, CA 94305; e-mail: . Received for publication: 06/16/04; accepted for publication: 10/20/04. 121 Journal of the American Medical Informatics Association Volume 12 Number 2 Mar / Apr 2005