The Genom of Homo sapiens.pdf

276 JAILLON ET AL.Detection of Evolutionary ConservedSequences (Ecores)The methodology we are currently using is based oncomparisons of the translated phases between two DNAsequences using TBLASTX (Altschul et al. 1990) as anengine. Although TBLASTX computation generates ahuge number of sequence matches called HSPs (highscoring pairs), these can be categorized as true positives(matches that are located within coding exons) or falsepositives (located elsewhere).We showed that it was possible to minimize the backgroundof false-positive matches (Roest Crollius et al.2000) to about 1% by (1) applying extensive and stringentmasking of low-complexity repeats, and (2) filtering HSPsusing appropriate combinations of threshold values for sequenceidentity rate and length of sequence alignments.For a given length of sequence match, we use a specificpercentage of sequence identity. These settings were initiallydetermined as a broken line manually, i.e., the frontierbetween putative true positives and false positiveswas empirically approximated by a piecewise linear function,using reference annotations. Since the number ofproperly annotated genes is well above 1000 in severalgenomes, we are now using polynomial approximationsboth for practical reasons and to improve accuracy.Finally, sets of overlapping true positive matches areassembled, and the resulting genomic regions are calledecores. Although an ecore designates a pair of correspondingregions, one on the target genome and one onthe query genome, the pair of regions may be composedof HSPs that are not in strict correspondence. Dependingon the context, we will employ the term “ecore” indifferentlyfor the pair or for one of its components.An example of such an optimization is shown in Figure1 for a set of 1589 Arabidopsis thaliana genes that werecompared to the Syngenta (Goff et al. 2002) sequencedraft of the rice genome. The fraction of false positives inthe test set on the right part of the curve (alignmentsabove thresholds for length and sequence identity) is below0.2%. However, this high specificity is at the expenseof sensitivity which is limited to ~64% of exons of the setof tested genes. These settings have to be defined for eachpair of genomes compared.Detection of Conserved Contiguity ofEcores (Ecotigs)Sequence conservation between genomic sequences ofspecies that have diverged for 100 Myr or more is essentiallyrestricted to coding segments. In a large majority ofcases, there is only a single ecore match in exons that arematched. In addition, if two ecores remain consecutive(contiguous) in both genomes A and B, they are highlylikely to belong to the same gene. In order to identifyecores that conserve such colinearity, named hereafterecotigs (ECOres conTIGuous), we designed an algorithmthat identifies conserved contiguity of ecores betweentwo genomes A and B by constructing well-chosen pathsin the joint “contiguity-similarity” graph. ContiguityFigure 1. The conditions that generate optimal alignment incoding regions were first tested using a large range ofTBLASTX (Altschul et al. 1990) conditions (W, X, scoring matrix)between a well-annotated set of 1589 genes including introns,exons, and 100 bp of intergenic region at both ends ofeach gene (P. Rouzé and S. Aubourg , pers. comm.) and the Syngentarice draft sequence (Goff et al. 2002). The BLAST settingsproviding the highest sensitivity are: match score = 15, mismatchscore = –3, W = 4, X = 13. All sequences were maskedagainst known repeats from rice and Arabidopsis. For each condition,a filter was applied based on the length and percent identityof alignments. The gray dots correspond to HSPs that overlappedexons in 100% of cases. The circles correspond to HSPsthat overlapped exons in less than 100% of cases. HSPs situatedto the left of the curve were discarded.edges were induced by exon neigborhood relationships ofecores from both genomes and similarity edges by correspondencebetween ecores (J.M. Aury et al., in prep.).Figure 2 outlines the rationale of the procedure we useto identify and construct ecotigs. Ecotigs group ecores togetheras long as colinearity is preserved, up to a fixed tolerated“gap” in ecore succession. In other words, ecoresthat are consecutive in genome A will be included in anecotig if they consist of at least two HSPs that are consecutiveor separated by one (the chosen gap value) additionalHSP at most on genome B (distance 1 and 2, respectively,in Fig. 2). When constructing ecotigs on genome A, wefirst consider the ecore pairs i and ii that are consecutiveon genome A (ia and iia). Ecore ia is linked to ecores ib1and ib2 on genome B. These latter are separated by a distance≤2 from ecore iib. According to the chosen rules, iand ii will be grouped in the same ecotig. The ecotig isthen tentatively extended to ecore iii. The distance separatingiib and iiib on genome B is 2. Consequently, iii willbe incorporated into the existing ecotig. By iterating of theprocess, we can group ecores i, ii, iii, iv, and v in a singleecotig. Ecores v and vi are contiguous on genome A (vaand via) but are not syntenic on genome B, and vb and vibare hence separated by an infinite distance. This will stopextension of the ecotig. A new tentative ecotig is then initiatedwith ecore vi and the process is iterated.A single ecotig on one genome may be related to multipleecotigs on the other one. In addition, ecotigs may becomposed of sequences from more than one gene, since