Molecular Databases Grow, and Grow, ... and Grow

Erica P. Johnson In the early days of American television, the late Steve Allen created a comic persona called "Question Man." An expert in whatever, Question Man contended that the public was so well-informed, knowing more answers than questions, that it was leaving many answers unquestioned. So, Question Man invited his audience to send him answers out-of-context, and he provided the questions. They were the punch lines. Today the soaring medium is the Internet, where hundreds of molecular databases hold answers to countless biological questions, most of which have yet to be asked, or even conceived. Increasingly popular among biologists, the databases are as challenging as they are invaluable. Information overload is but one hurdle: Every day, laboratories contribute new information that must be vetted, organized, and made accessible to life scientists who are not trained in computer science. Database ResourcesClick for larger version (PDF, 37K) "The biomedical and biological research community is the largest research community on the planet," says Ewan Birney, team leader for genomic annotation for the EMBL (European Molecular Biology Laboratory) Nucleotide Sequence Database. "And it goes all the way from a research clinician who is working on humans to an evolutionary biologist working on shrimp that are inside of ice in the Antarctic. And all those people want to access our data for one reason or another." In his 2003 update of the Molecular Database Collection,1 Andreas D. Baxevanis, of the National Human Genome Research Institute, tallied nearly 400 key databases--everything from the widely used UCSC (University of California, Santa Clara) Human Genome Browser and the Swiss Protein Knowledgebase, to specialized sites such as the Olfactory Receptor Database and the Lipase Engineering Database. Bioinformaticians who maintain the bigger databases say these databases are underused, despite heavy traffic. For example, 250,000 users perform 100,000 BLAST searches a day at the National Center for Biotechnology Information (NCBI), and 4,000 users request 100,000 pages a day at the UCSC browser. "Most scientists don't have a good idea of the breadth of information available to them," says Baxevanis. Besides basic sequence information, these databases often provide data on gene expression, intermolecular interactions, metabolic pathways, pathology, biological function, and more. If a scientist does not know all the possibilities, a search that seems successful might not tap into all the data that is available online, according to David Wheeler, head of NCBI's user service division. "A lot of biologists are motivated to go to a resource only when they need something," he says. "Maybe they find [what they are looking for], and maybe they only think they found it." Wheeler urges scientists to browse NCBI's collection of databases as well as other sites to see for themselves what's online. STANDARDS ISSUE One issue that quickly becomes apparent is standardization. Nomenclature, a perpetual problem in biology, is complicated both by different practices among scientific communities and by variations in database form, content, and user interface. "Right now the mouse and human and rat nomenclature committees work closely together, but we don't always have the same nomenclature for the same gene," says Judith A. Blake, a principal investigator with the Mouse Genome Informatics group at the Jackson Laboratory, Bar Harbor, Maine. The group integrates the data, thus enabling scientists to ask for complete information, such as all genes mapped on chromosome 2 that have human orthologs. If the user is searching for a gene name that is not in the database, however, a search might turn up zilch even though the information is there. Blake cites the Kcnj6 gene. Originally known as weaver, it has six synonyms in the database, all of which point to the same public accession number (MGI:104781). "It would be great if people put accession IDs in scientific papers," says Blake, "but it's not standard at this point." One solution to the nomenclature problem would be to speak with a common language. Gene Ontology (www.geneontology.org), for instance, a defined, hierarchical vocabulary that databases can use to circumvent the language problem, defines terms for molecular functions, biological processes, and cellular components. Other vocabularies also exist, including Plant Ontology (www.plantontology.org), and Sequence Ontology (www.gmod.org/so.shtml). NCBI is tackling another standardization problem with its Reference Sequence project. GenBank, NCBI's DNA database, can have 100 different sequences entered for the same gene, according to Wheeler. The project seeks to weed out mutations, sequencing errors, and cloning artifacts to provide a validated and consistent baseline sequence. But this begs the question: Just how accurate is the data? ANNOTATIONS: CAVEAT EMPTOR While Baxevanis expresses confidence that sequencing data from the Human Genome Project is 99.9% accurate, he and Wheeler both say people need to be careful about the annotation. "You need to maintain a healthy sense of skepticism," suggests W. James Kent, who helped develop the UCSC browser. Accuracy issues relate to how the information that is entered into molecular databases is generated. The job of data entry falls to the database curators. Many annotations result from gene- prediction software, which, though steadily improving, remains far from perfect. The curators attempt to reconcile these predictions by searching for errors, evidence, and inconsistencies in the experimental data found in published reports or submitted by scientists. It is a slow, labor-intensive, and expensive process. "Cleaning up this sort of stuff occupies a good deal of our time," says Kent. "Figuring out what data you can and cannot trust is much of the work of a bioinformatician." The UCSC browser routinely identifies the source of its data. Not all databases do, however, and Birney describes maintenance of evidence trails as a problem that urgently needs to be addressed.2 When errors are discovered or annotations change, no automatic process exists to update all of the references to that information. As a result, out-of-date and incorrect annotations abound. "It's just a headache, one of a number of headaches that [curators of] these databases have to cope with," Birney says. Despite such foibles, database developers say accuracy issues are more troubling to computer scientists, who expect calculations to produce precise answers automatically, than to biologists, who are accustomed to working with contradictions in data. Kent chooses to post multiple lines of evidence that don't agree and let the users address the contradictions through more research. If 10% of data turns out to be false genes, he says the biologist is better off than missing a gene. "This is research. We never know the right path until we do," says Baxevanis. LOG ON, QUERY, REPEAT Online databases update continually, so repeating searches is generally recommended. Birney and Kent suggest every three months; Wheeler and Baxevanis suggest every 30 days. "It depends on how quickly the field you are in is moving. It might be good to come every week," Wheeler adds, noting that changes tend to be abrupt rather than gradual. Someone has made a breakthrough, and suddenly the data is available. "One day what you want is not going to be there at all, and the next day it is there entirely," he says. Another option is to use the services of a private company that integrates and curates data. Biomax Informatics in Martinsried, Germany, integrates public data with 4,000 genes identified by its own proprietary system into one 37,000-gene (and growing) human genome database, according to Christine Schuller, manager of scientific projects. She says it also has 240 other genomes, and can produce a new database for its customers. Biosift in Cambridge, Mass., integrates 600 genomes into a single interface that makes comparisons across sites easier for its customers. "Sometimes it's the same data ... mapped out by different groups. And so you can see the divergence between the groups," says marketing and communications director Viviane Siino. About 5,000 customers use the Celera Discovery System from Applied Biosystems of Foster City, Calif., according to senior director and applications developer Matteo di Tommaso. In addition to integrating the company's proprietary human genome data with public databases and making the information easier to use, di Tommaso says the group is developing assays-on-demand for scientists who want to do "wet lab" experiments based on in silico findings. "We want to make the circle from hypotheses to experiment to data interpretation more rapid," he says. "Customers can click through from gene to assays that interest them." With so much data in the public domain, do life scientists really need these private services? "There's more than enough data publicly available to occupy yourself," says Wheeler. Baxevanis says scientists have to decide for themselves whether the added information is of value or not. The market for the private competitors will not be in proprietary genes, but in services, according to Birney. "The most successful companies are the ones that add value, and those end up becoming quite [service oriented]--consultancy style companies--rather than, 'We have some special things that you don't have. Buy our stuff,'" he says. Integration will happen, according to many of the people waging, in Birney's words, an "endless fight against chaos," as they develop and maintain the databases. "I think it will always be a network of databases, where certain data sets become bigger and richer and absorb other data sets," he says. "And for all the absorption that's happening by well-established systems, you'll be getting new data sets being built up." Wheeler foresees a more stable phase when data is centralized in a small number of large systems such as the NCBI. The big challenge, says Kent, will be finding better ways to assign function to genes. A more immediate question for many life scientists is whether they need to hire a bioinformatician. Unless a lab is heavily invested in problems that require intensive use of bioinformatics, the database builders say the answer is, probably not yet. For now, Birney suggests looking for "a geeky graduate student who is up to the job," or three or four labs can get together to share a scientist who has bioinformatics expertise. That person will need to learn Perl, Java, or Python (programming languages that are easier to use than C++) when setting up a lab database or linking lab data to a large database for internal use. Says Birney, "I think most of molecular biology is going to have to get used to the fact that rather than having a lab notebook, you really should have a lab database. Over the next 10 to 20 years, I think we'll see that transition." Jane Salodof MacNeil (macneiljs@earthlink.net) is a freelance writer in Groveland, Mass. References 1. A.D. Baxevanis, "The Molecular Biology Database Collection: 2003 update," Nucleic Acids Res, 31:1-12, Jan. 1, 2003. Note: The article introduces an annual review in which developers of leading databases report on their resources; available at nar.oupjournals.org 2. E. Birney et al., "Databases and tools for browsing genomes," Annu Rev Genomics Hum Genet, 3:293-310, 2002. TWO VIEWS FROM DATABASE USERS Bin Liu: "Democratizing Information" A senior research scientist and bioinformatician, Bin Liu studies toxicology at Pfizer in Groton, Conn. He uses databases virtually every day, both in his own work and for colleagues who seek help with data mining. Among the public databases, he most often frequents those of the National Center for Biotechnology Information (NCBI); "I think PubMed is unparalleled," he says of NCBI's archive of life sciences journals. He also has access through company subscriptions to proprietary databases such as the Celera Discovery System and ToxExpress Solutions, a product of Gene Logic in Gaithersburg, Md. The public databases tend to be broader, whereas the proprietary databases are better curated and easier to search, says Liu. If he is looking for targeted information--for example, what tamoxifen will do to a rat's liver--he goes to a private toxicology database. The information he seeks may be available in public databases, but Liu says these are more cumbersome to search. For him, deciding which database to use is often a trade-off between time available and the benefits of exploring. "It's too much information," he says. "You basically choose the ones that you're most comfortable with, and you use them often." Liu credits databases with "democratizing information" by fostering an open flow that is significant for his generation of biologists. Referring to the 50th anniversary of the discovery of the double helix, he observes: "Watson and Crick's discovery, in large part, owes to their first-hand access to Rosalind [Franklin]'s X-ray data. Imagine, then, all the data would be wiped instantly to a database and accessible to everybody. Guess what, the game play could be quite different." Yong Cang: "Life is too short" Yong Cang is attempting to knock out mouse genes for signal transduction studies at Columbia University's College of Physicians and Surgeons. His laboratory in the Department of Biochemistry and Molecular Biophysics has a subscription to Celera Discovery System, but he also uses public databases that focus on the mouse genome. Cang says he goes to them only for specific pieces of information, however, and that he usually asks "real people in the field" which one to use before he searches. "Life is too short," he says, describing an extended visit to the NCBI web site as unproductive: "I spent days browsing the databases, and I got confused. Based on my prior experience I won't waste time in it. Labs doing a lot of bioinformatics, they may have the expertise, but we don't have the manpower, and we don't have the time to learn." Accuracy is also a big concern for Cang. Whenever he retrieves a sequence from a database, he locates the gene manually to verify the information. "I verify all the information for myself. I never simply trust the database or the published stuff," he says. While Cang sees bioinformatics as "potentially very innovative," he expresses concern that its role has been overstated. "There must be a well-balanced mixture of basic research and bioinformatics," he says. "Bioinformatics can do a lot of things trying to give life to the letters in the database, but most of the time bioinformatics won't tell you the results." function sendData() { document.frm.pathName.value = location.pathname; result = false if (document.frm.score[0].checked) result = true; if (document.frm.score[1].checked) result = true; if (document.frm.score[2].checked) result = true; if (document.frm.score[3].checked) result = true; if (document.frm.score[4].checked) result = true; if (!result) alert("Please select") return result; } .myradio{background-color : #94bee5} Please indicate on a 1 - 5 scale how strongly you would recommend this article to your colleagues? Not recommended 1 2 3 4 5 Highly recommended Please register your vote

Molecular Databases Grow, and Grow, ... and Grow

Interested in reading more?

Become a Member of