The Genom of Homo sapiens.pdf

Ensembl: A Genome InfrastructureE. BIRNEY AND THE ENSEMBL TEAMEBI, Wellcome Trust Genome Campus, Hinxton, Cambridge CB10 1SD; and Sanger Institute,Wellcome Trust Genome Campus, Hinxton, Cambridge CB10 1SA, United KingdomThe genome sequence of any organism is an invaluableresource for molecular biologists. Experiments are eithertrivial to design or greatly enhanced by the knowledge ofthe genome sequence; linkage analysis leads directly to aset of genes in the critical region, and association studiescan be designed to any region (including, in theory, theentire genome). At least as important as the ease of experimentaldesign is the fact that the genome is essentiallycomplete, and therefore, researchers can be confidentthat the aspects of biology they are studying must bepresent in the genome sequence in some manner. The factthat all biology is somehow associated with some aspectof the genome, that the genome is complete and essentiallyunchanging, means that it is a foundational resourcefor biology. The generation of the human (Landfer et al.2001) and mouse (Waterston et al. 2002) genome sequencesprovided landmarks for the understanding of humanbiology.There is a catch. The genome sequence of large organismsis itself a large, unwieldy data set and is opaque toanalysis: For all of the arguments of completeness of thegenome, if we can’t ascribe at least some function to partsof the sequence, it becomes effectively unusable. In additionto this scientific problem of knowing which sequencesare functional or not, there is the somewhat mundaneproblem of simply handling the data size. Thisengineering problem is compounded due to the number ofgenomes now sequenced and the churn rate of genomeand cDNA sequence.Ensembl is designed to overcome these problems andtherefore to make genomes far more useful to a broadrange of audiences (Clamp et al. 2003). Ensembl focuseson providing information to three classes of users: (1)Bench biologists, who generally are focused on one ortwo genes and want a user-friendly, graphical web-basedsystem to access the genome. The ensembl web site,www.ensembl.org, is focused on this user. (2) Mid-scalefunctional genomics users, who are working with sets ofgenes, either due to positional cloning or expression analysis.These users often need their own “slice and dice”data-extraction routines. The EnsMart data-mining system(described below) is focused on this user. (3) Largescalegenomics groups and other bioinformatics groups.We have found that by simply being open in terms of bothsoftware and data, we are able to satisfy most of thisgroup’s needs.Ensembl is not the only group analyzing and displayingthese genomes. The UCSC group under David Hausslerand the NCBI group are both very active in this area.We enjoy healthy competition with these groups and collaborateon the underlying data resources, ensuring, forexample, that the assembly is common between all threesites.RESULTSTable 1 outlines the genomes which Ensembl displays.We make a distinction between genomes where we predictgenes and genomes where the gene structures areprovided by another group. Notice that for all genomesthere have been multiple gene builds, in each case marshalinga set of data resources (e.g., cDNA and EST datasets) and tuning the gene prediction process for eachgenome. A gene build takes information from threesources. cDNAs generated from the target genome arereconciled back onto the genome sequence by a specific“best in genome” process. Then, remaining pieces of thegenome which show strong protein similarity to genes inother organsisms are used to generate “novel” genes viathe program genewise. Finally, EST sequences aremapped back to the genome, clustered, and then a minimumset of transcripts which represent the clusteredESTs are generated. Depending on the species, the ESTdata are sometimes merged with the main cDNA and proteinbuild (e.g., in Anopheles gambiae) and sometimes arekept separate (e.g., in Homo sapiens). This is due to thedifference in EST quality in different genomes, with thelarge, error-prone human EST set proving the hardest touse.The Ensembl web site (www.ensembl.org) is designedfor biological researchers to quickly orient themselves onthe genome and design experiments on the basis of thegenome information. Our two main displays are focusedon genomic sequence and gene products. Increasingly,we have discovered that more and more people want towork with subsets of gene products encoded by a particulargenome. This “set” working behavior is catered to bythe EnsMart data-mining tool available both through theWeb and as a downloadable command-line tool.Assessing the quality of our gene prediction is hard becausewe use all available data at any point in our buildprocess, and new experimental cDNA approaches tend tofocus on currently undiscovered cases. Assessments viaoverlap to dual-genome predictors, which use only thehuman and mouse genome as input (and therefore shouldbe unbiased toward other cDNA or EST evidence), suggestthat there are around another 1,000 protein-codinggenes outside of Ensembl to predict (Guigó et al. 2003).Cold Spring Harbor Symposia on Quantitative Biology, Volume LXVIII. © 2003 Cold Spring Harbor Laboratory Press 0-87969-709-1/04. 213