Ensembl Compara - CNB - Protein Design Group

Timeline 2 : Genomes9/01 Release with Mouse genome (gp)2/03 Ensembl 10 (Fugu, worm, fly, Anopheles, briggsae; 1stmajor schema/API rewrite)11/03 First Ensembl SAB1/04 Ensembl 20 (2nd major schema/API rewrite)12/04 Ensembl 27 (Variation schema/API rewrite)2/05 Ensembl 28 (archive sites, bioMart replaces Mart)7/05 Ensembl 32 (website redesign)10/05 Ensembl 34 (multiple genome alignment data/AlignSliceAPI)10/05 WT Funding approved (WTSI + Ensembl) until 201112/05 Switch from monthly to bi­monthly cycle2/06 Ensembl 37 (mouse SNPs, transcriptSNPview)4/06 Ensembl 38 (Combined Havana­Ensembl Humangeneset)12/06 Ensembl 42 (Platypus!)4/07 Ensembl 44 (37 species, more than 500 Gb behind thewebsite)Jul 2007­ T. Hubbard

Ensembl: What do you get?Jul 2007Genome Annotation– Protein coding gene structureConsistent with genome, predicted across all vertebratesManual annotations (human, mouse, zebrafish, MHC)– RNA genes (including miRNA)• Consistent with genome, predicted in across mammalsAdditional identifiers per genesAffymetrix, EntrezGene, Uniprot…• Comparative & Functional GenomicsGenome alignmentsBlastz, Blat, coordinated with UCSC– Homologues between genomes– Protein treesVariants (SNPs), strains, genotypesTiling array data• Infrastructure– Website, Data mining tool, database and data dump– Portable, extendable, open source system with database, API, website,pipeline

Ensembl groupsJul 2007GeneBuilders: sequence masking, gene buildingCore: database schema, stable id mappingCompara: protein homology, genomic sequencealignmentsFunctional Genomics: SNPs, probe mapping,functional dataMart: martification of the previous data for dataminingWeb: web site, new views for new dataUser support: help, workshops, tutorialsMore people: Research, DAS, VectorBase, Zebrafish,Systems...

*Ensembl release cyclegenebuilder (10+3+4)~ 3 monthscore (4)compara (6)functional genomics (5)mart (4)web (7)release!!1 release coordinator+ 1 assistantJul 2007

BilateriaNCBI Taxonomy for the 31species in Ensembl 43ChordatesAmniotesTetrapodsMammalsEutherianPrimatesGliresCarnivoraHomo sapiensPan troglodytesMacaca mulattaRattus norvegicusMus musculusCavia porcellusOryctolagus cuniculusTupaia belangeriBos taurusErinaceus europaeusFelis catusCanis familiarisLoxodonta africanaEchinops telfairiDasypus novemcinctusMonodelphis domesticaOrnithorhynchus anatinusGallus gallusXenopus tropicalisDanio rerioTakifugu rubripesTetraodon nigroviridisGasterosteus aculeatusOryzias latipesCiona intestinalisCiona savignyiAedes aegyptiAnopheles gambiaeDrosophila melanogasterCaenorhabditis elegansSaccharomyces cerevisiaeJul 2007

Genomes: the futureAnother 28 mammals being sequencedMain driver is evolution, but perhapsopening up biology• New technology sequencing454 ­ currently about 5 fold cheaperSolexa ­ currently about 100 foldcheaperJul 2007

Ensembl ComparaA single database which contains precalculatedcomparative genomics data and which is linked toall the Ensembl Species databases.Access via web interface, perl API and mysqlA production system for generating that databaseJul 2007

Compara dataRaw genomic sequenceWhole genome alignments(tBLAT, BlastZ­net, PECAN)Syntenic regions (based on BlastZ­net)Protein sequence Raw protein alignments (wublastp) Protein Family clusters Protein trees (since Jun 2006 ­ v39) Gene orthology / paralogy predictionsJul 2007

Genomic AlignmentsBlastZ­Netused to compare closely related pair of speciesBlastZ­raw ­> BlastZ­chain ­> BlastZ­netJul 2007Translated BLAT used to compare more distant pair of speciesPecanmultiple global alignments all vs all coding exons wublastp ­> Mercator ­>Pecan on each syntenic block

Global vs. Local AlignmentsGlobalLocalinversionduplication1 21 2(­)LocalAdvantagesFor large genomic regionsCan identify inversionsDisadvantagesFails to identify insertions ordeletionsGlobalCan detect insertions ordeletionsFails to detect inversionsJul 2007

Pecana consistency based multiple­alignment programATGGGCTTTTGCATTTG}ATGGGCAGCATTTGACGGGCATTTGCTTCTGATGGGCTTTTGCATTTGATGGGC­­­AGCATTTGvsATGGGCTTTTGCATTTGATGGGCA­­­GCATTTGATGGGCTTTTGCATTTGACGGGCATTTGCTTCTGJul 2007ATGGGCA­­­GCATTTGACGGGCATTTGCTTCTGTakes into account all pairwisealignments, across the entire treeATGGGCTTTTGCATTTGATGGGCA­­­GCATTTGACGGGCATTTGCTTCTG

Pecan optimizationscut-lineLook for anchors (exonerate)perform a banded alignmentBreak­up alignments intofragmentsMuch redundancy betweenpairwise alignments: usetransitive anchorsBhiddencut-pointACABJul 2007C

Encode ComparisonSPECIFICITYCOVERAGEJul 2007

Jul 2007Example of AlignSliceView betweenHuman/Mouse/Rat/Dog with PECAN

How to align suchlarge segments?New challengesHow to deal withduplications?Jul 2007How to predictancestralgenomes?

EnredoAnchors500.000 anchorsfor mammals­­morethan 1 anchorper 10KbJul 2007

EnredoSolving the graphJul 2007Common paths defineco­linear regionsThis process allows for mismatches in the paths

Jul 2007ORTHEUSan ancestral sequence inference programAddresses the inference of insertion­deletion historiesand substitution eventsStarts from a (multiple) alignment and assumes afixed treeReconstructs the ancestral sequences in the tree andrefines the input alignmentInsertion/deletion events are handled using a branchtransducer modelSubstitution are handled using Tamura­Neinucleotide substitution modelAncestral sequence are inferred using weightedsequence graph

ORTHEUSBranch Transducer ModelFour transition parameters model:Insertion(cont.)Applied to a tree:Deletion(cont.)InsertionDeletionJul 2007Matches and insertion/deletion eventsare propagated down the tree

Protein Homology (e! 38):Orthologue predictions based on ‘best reciprocalblast hits’ and ‘synteny extensions’Young paralogues for a selected set of species e! 39+:orthologues and paralogues are inferred fromprotein treesJul 2007

HomologyBRH/RHS ­­ pairs of species (n*(n­1)/2)Cases of fast evolving genesGlobal view of the evolution history of the gene considered✔Phylogeny: Orthology/Paralogy in one goJul 2007

Load Genes and LongestTranslation from all species inEnsembl. 25 species in v40hoursWU Blastp + SmithWaterman longesttranslation of every Gene against everyother in a genome­wise mannerhours/daysBuild a graph of gene relationsbased on BRH 1 and BSR 2Extract the connected components(= single linkage clusters)2­3 days ­ 1CPUFor each cluster, build a multiplealignment (MUSCLE) based on theprotein sequenceshoursFrom each alignment, build a genetree (PHYML)hours/daysReconcile each gene tree with thespecies tree to call duplicationevent on internal nodes (RAP)hoursInference of orthologs and paralogs(OrthoTree)hoursJul 2007BSR: Blast Score Ratio. When 2 proteins P1 and P2 are compared,BSR=scoreP1P2/max(self­scoreP1 or self­scoreP2). The default threshold usedin the initial clustering step is 0.33.

Load Genes and LongestTranslation from all species inEnsembl. 26 species in v41hoursWU Blastp + SmithWaterman longesttranslation of every Gene against everyother in a genome­wise mannerhours/daysBuild a graph of gene relationsbased on BRH 1 and BSR 2Extract the connected components(= single linkage clusters)2­3 days ­ 1CPUFor each cluster, build a multiplealignment (MUSCLE) based on theprotein sequenceshoursmultifurcated species treeReconcile each gene tree with thespecies tree to call duplicationevent on internal nodes (RAP)hoursv41From each alignment, build a genetree (PHYML)hours/daysInference of orthologs and paralogs(OrthoTree)hoursJul 2007BSR: Blast Score Ratio. When 2 proteins P1 and P2 are compared,BSR=scoreP1P2/max(self­scoreP1 or self­scoreP2). The default threshold usedin the initial clustering step is 0.33.

Load Genes and LongestTranslation from all species inEnsembl. 31? species in v42WU Blastp + SmithWaterman longesttranslation of every Gene against everyother in a genome­wise mannerBuild a graph of gene relationsbased on BRH 1 and BSR 2Extract the connected components(= single linkage clusters)For each cluster, build a multiplealignment (MUSCLE) based on theprotein sequencesInference of orthologs and paralogs(OrthoTree)From each alignment, build a geneand reconcile with the species treeto call duplication events oninternal nodes (NJTREE_PHYML)v42Jul 2007BSR: Blast Score Ratio. When 2 proteins P1 and P2 are compared,BSR=scoreP1P2/max(self­scoreP1 or self­scoreP2). The default threshold usedin the initial clustering step is 0.33.

Protein trees and homologyDuplication nodeSpeciation nodeH1M1H2ortholog_one2oneH1:M1between_species_paralogM1:H2EuarchontogliresH2’M2M2’within_species_paralogH2:H2’ Homo sapiensortholog_many2manyH2:M2, H2:M2’,H2’:M2, H2’:M2’within_species_paralogM2:M2’ Mus musculusH3M3ortholog_one2manyH3:M3, H3:M3’within_species_paralogM2’:M3’ EuarchontogliresM3’Jul 2007Orthologues : any gene pairwise relation where the ancestor node is a SPECIATION event.Paralogues : any gene pairwise relation where the ancestor node is a DUPLICATION event.

A special case of homologyDuplication nodeSpeciation nodeMortholog_one2oneH:Mgene lossR’HH’gene lossgene lossM’apparent_ortholog_one2oneM:R, H:RRJul 2007Orthologues : any gene pairwise relation where the ancestor node is a SPECIATION event.Paralogues : any gene pairwise relation where the ancestor node is a DUPLICATION event.

Gene tree : 1st data assessmentGood concordance with the classical BRH/RHS pairedspecies approach (RHS are based on gene order conservation)Find more complex one­to­many and many­to­many relationsHuman/MouseHuman/DrosophilaRHSBRHNEWBRHNEWmany2many1771131,439many2many1701,599one2manyone2one7252051,30910,7362,81510919,381one2manyone2one1,8708804,5638011,443apparent one2one781,571104apparent one2one2,040241lost2,0272,060lost620Future plans: convergence with TreeFamJul 2007

FamilyGene family clustering predictionsRuns on all Ensembl transcripts plus allUniprot/SWISSPROT and Uniprot/SPTREMBLmetazoan proteinsAlgorithm is based on all vs all blastp, MCLclustering, Muscle multiple alignerJul 2007

Jul 2007