Reading the Human Genome

To translate the billions of sequenced bases of the human genome into meaningful information, investigators in the private and public sectors continue to use computer prediction to automatically annotate the human genome--to attach biological footnotes about genes, the proteins they encode, and the ailments to which they may be linked. But as with any translation, there's potential for multiple interpretations. Annotation projects could present geneticists with significantly different genome schematics. "I think that for the moment scientists are going to have to revel in the variety," maintains Robert Waterston, director of the genome sequencing center at Washington University in St. Louis. A joint effort in the United Kingdom from the European Bioinformatics Institute (EBI) and the Sanger Centre, U.S. efforts at the National Center for Biotechnology Information (NCBI) and Oak Ridge National Laboratory (ORNL), and commercial efforts will present different annotations based on the data available, algorithms employed, and weight given to different data points. Mass confusion isn't likely, because most scientists working with sequence understand the caveats of annotation. But supporting information on the methods and programs involved will be essential. "If you are making assertions about features of the genome, then it needs to be very clear on what evidence you make that assertion," remarks Graham Cameron, joint head of the EBI, an outstation of the European Molecular Biological Laboratory. Automatic annotation uses computers to sift through the huge amount of Human Genome Project sequence information and predict biological features based on similar, previously known genes belonging to humans or other organisms. Investigators also rely on manual annotation, especially when a gene has been well characterized and its complementary DNA (cDNA)--genomic DNA sequences based on RNA templates--has been cloned. "Obviously you'd want to distinguish that kind of situation, where you really had good experimental evidence to back up the annotation, from one where it was purely computer prediction," comments Philip Green, a professor of molecular biotechnology and bioinformatics specialist at the University of Washington in Seattle. According to John Bouck, director of informatics at the Baylor College of Medicine Human Genome Sequencing Center, the uncertainty inherent in automatic annotation has made those annotations somewhat more conservative. Says Bouck, "We don't want to be misleading people and saying, 'Hey, there's a gene here that's a kinase' when in fact it's not." Multiple Efforts, Multiple Annotations As much an art form as a science, all annotation projects and the associated software will be continually updated and refined for years to come. The ORNL and its collaborators in the Genome Annotation Consortium, which includes the University of Pennsylvania Center for Bioinformatics, the Baylor College of Medicine, and an international collaboration called the Genome Database, were the first to get into the human genome automatic annotation game about three years ago. They continue to annotate the human, mouse, and 30 microbial genomes, and they have several annotation-related servers online. The EBI-Sanger Centre project, called Ensembl, and the NCBI stepped up annotation efforts in the last year. Ensembl recently made its first annotation of a large portion of the human genome available online. To make their respective annotation databases more comparable, scientists with Ensembl, the NCBI, and ORNL have agreed to apply their automatic methods to the same reference DNA sequence. Like Ensembl and ORNL, NCBI will make its annotation completely available to the public. Not yet as well organized as sequencing efforts, annotation collaborations between publicly funded projects continue to evolve. Ed Uberbacher, head of computational biosciences at ORNL, expresses disappointment with the communication between NCBI and ORNL. Says Uberbacher in reference to the NCBI, "It would've been nice in an endeavor as important as this one if a greater level of cooperation and collaboration could've been reached." His group is in discussions with the Ensembl group about a possible collaboration. Commercial ventures also have a great deal invested in annotation projects. Rockville, Md.-based Celera Genomics Group reportedly plans to do a comprehensive automatic annotation--to be available on a subscription basis--based on its own sequence data and that from the public project. (No representative of Celera could be reached for comment on this story.) Companies such as Incyte Genomics, Palo Alto, Calif., are doing more targeted annotation of genes that may, for instance, be linked to diseases. Doubletwist, of Oakland, Calif., which announced the first extensive annotation of the human genome last May, contends that its automatic annotation, also available via subscription, will be more comprehensive than public efforts despite its only using the publicly available Human Genome Project sequence. "What we're trying to do is use multiple algorithms and multiple methodologies and then synthesizing," comments Doubletwist director of research Nicholas Tsinoremas. Cameron, though, points out that Ensembl also takes advantage of several algorithms as well as confirmatory information from databases. Ensembl, a joint undertaking by the European Bioinformatics Institute and the Sanger Centre, and efforts from the Oak Ridge National Laboratory and its Genome Annotation Consortium are two of several human genome annotation projects currently underway. Indeed, there's no obviously superior annotation method or methods, and it's not clear whether the annotation done by Doubletwist (or even, in the long run, that done by Celera) will be significantly better than what's freely available from Ensembl, ORNL, and NCBI. According to Bouck, though, commercial annotations may be better suited to specific questions about specific diseases. A researcher interested in Parkinson's disease, for instance, might be able to more effectively reconcile the literature and the sequence and experimental data to identify regions of interest. "It's actually probably better when [annotations] do disagree," comments Green. "Then the scientist can say, 'Aha, we don't know how the gene is delineated here so we better do some more experiments to find out.'" Waterston suggests that because there's so much data to be processed, database interface and ease of use, rather than the sequence annotation itself, may end up differentiating between, say, Celera's annotation project and NCBI's or Ensembl's. Annotation Challenges No matter who's doing the annotating, filtering out the genomic noise and finding the meaningful bits of the human genome figures to be difficult. The main goal of automatic annotation efforts right now: singling out genes--believed to number between 30,000 and 100,000--from a sea of sequence. Prior to doing so, however, scientists must identify repetitive elements in the genome, repeated sequences that Green likens to viruses because they use the cell's machinery to multiply and reinsert themselves into the genome. At least half of the genome is believed to consist of these repeat families. Repetitive elements haven't posed a big problem for annotators, however. A program that Green and bioinformatics expert Arian Smit helped develop, called Repeat Masker, has proven effective for weeding out this type of so-called junk DNA. Much more difficult to deal with, and much less understood, is the phenomenon of alternative splicing. Alternative splicing occurs when the same gene encodes two or more different proteins because the gene's associated RNA is spliced several different ways. Bouck estimates that alternative splicing affects almost 40 percent of the transcripts of the human genome. Even if annotators are lucky enough to identify a gene with relative certainty, they still might be perplexed as to exactly how that gene affects biological function. It can get exceedingly complex when considering protein function--how proteins interact, in which tissues they're expressed, and under what conditions. "If we really want to understand the biology of the human, then we have to be able to understand that question," says Waterston. Celera and several academic groups are currently working on a sequenced and annotated mouse, which should help provide the answers. In general, if there's a sequence in the mouse that's homologous to one in the human, then it's a good bet that that sequence has some role in the human, a comparison that helps researchers do human annotation. Because most coding regions between the human and mouse are conserved at a level of about 70 percent, researchers should be able to use data from the mouse genome to identify alternatively spliced human exons. Determining protein function still won't be easy, though. "It turns back into a large biological experiment," says Bouck. Waterston also points out the value of knowing the mouse genome to better understand the mouse genes themselves--a task that will require much higher-quality sequence--which will then allow researchers to do highly sensitive comparisons between the mouse and human. But to comprehensively compare the genetic makeup of the mouse, human, and other species, and to accurately annotate the human genome in the process, researchers may have to do more than refine computer methods and sequence DNA. They'll likely have to come up with a common nomenclature for genes and their products--precisely the goal of a recently formed group of scientists called the Gene Ontology (GO) Consortium. Claiming that the labeling of shared biological elements has not kept up with sequencing efforts, researchers working on fruit fly, mouse, and yeast genomes plan to produce a dynamic, controlled standard vocabulary for several organisms. They hope to describe the elements of biological processes, molecular functions, and cellular location in all eukaryotes--even as understanding of gene and protein roles continues to change.1 "The idea is to have a limited vocabulary on which computers can 'A,' rely, and 'B,' compute," explains David Botstein, chairman of genetics at Stanford University and participant in the project. GO members want to make clear, for instance, if and when researchers are talking about the same functional proteins in different organisms even if those proteins are engaged in different processes. "The really essential advance has nothing to do with computers," remarks Botstein. "The sort of standard way of talking about gene function is impossibly vague and has to be made more specific if it's going to be useful." Says Bouck, "It's one thing to annotate the human and be able to see all the genes that are there. But really, what's going to be fun, and what's going to be interesting, and what's going to enlighten us, is looking at the mouse and looking at other organisms. If it's not possible to transition between organisms, the annotation problem is going to be just intractable." Eugene Russo can be contacted at erusso@the-scientist.com. 1. M. Ashburner et al., "Gene Ontology: tool for the unification of biology," Nature Genetics, 25:25-9, May 2000. For More Information Celera Genomics Group www.celera.com/corporate/prodandserv.cfm Doubletwist www.doubletwist.com Ensembl Project www.ensembl.org Gene Ontology Consortium www.geneontology.org National Center for Biotechnology Information www.ncbi.nlm.nih.gov Oak Ridge National Laboratory and Genome Annotation Consortium compbio.ornl.gov/gac

Reading the Human Genome

Interested in reading more?

Become a Member of