HOGENOM a phylogenomic database Simon Penel Pascal Calvat

HOGENOM a phylogenomic database Simon Penel, Pascal Calvat, Jean-Francois Dufayard, Vincent Daubin, Laurent Duret , Manolo Gouy, Dominique Guyot, Daniel Kahn, Vincent Miele, Vincent Navratil, Guy Perrie re, Re mi Planel

Several phylogenomic databases developed at LBBE/PRABI • HOVERGEN – Verterbrate Proteins from Uni. Prot – Clustering with Si. Li. X • HOMOLENS – Proteins from Ensembl Complete Genomes – Clustering from Ensembl – Trees calculated annoated (S, D, L) with new methods (Phyl. Dog, LBBE) • HOGENOM – – Proteins from all available complete genomes (Bacteria, Eukaroyota, Archaea) Clustering with Si. Li. X and post-processing with Hi. Fi. X Trees will be annotated (S, D, L, T)

HOGENOM characteristics • all complete genomes from the whole tree of life (not restricted to particular phylum) • Propose « gene families » : full length homologous sequences (different of « domain families » )

Domain vs. gene families Protein domain family Families of homologous protein domains (Pro. Dom): - Evolution by domain shuffling (duplication, loss, translocation)

Domain vs. gene families Homologous gene family Families of homologous protein domains (Pro. Dom): - Evolution by domain shuffling (duplication, loss, translocation) Homologous Gene families (HOGENOM): - Evolution of homologous genes by speciation or by gene duplication, or horizontal transfer - Sequences are homologous over their entire length (or almost)

Orthologs and paralogs in HOGENOM is centered on phylogenetic trees of gene families. Information on orthologs and paralogs can be deduced from gene trees: - from the annotation of gene trees (Duplication, Speciation, Transfer) - from query tools such as tree-pattern matching

Building Compare all proteins against each other (BLAST) Cluster homologous sequences into families (SILIX + HIFIX) Compute multiple alignments for each family Compute phylogenetic trees for each family Annotate phylogenetic trees (gene duplications, losses, transfers)

Compare all proteins against each other • Iterative BLAST calculation • Use of a non-redundant protein sequence database … (all know proteins , about 20, 000 non redondant sequences) … associated with a resulting BLAST hits database (from which blast hits may be extracted) • Cluster, grid and cloud computing

Building Compare all proteins against each other (BLAST) Cluster homologous sequences into families (SILIX + HIFIX) Compute multiple alignments for each family Compute phylogenetic trees for each family Annotate phylogenetic trees (gene duplications, losses, transfers)

Si. Li. X 1 st step : similarity search Protein database BLASTP BLOSUM 62 -4 E ≤ 10 Local pairwise alignments

Si. Li. X 2 nd step : Si. Li. X clustering Use the all-against-all BLAST hits

Si. Li. X : Selection of consistent HSPs Seq. A S 3 S 1 S 2 S 4 Seq. B S 1’ Seq. A S 2 Seq. B ∆lg 1 lg. HSP 1 ∆lg 2 lg. HSP 2 ∆lg 3

Si. Li. X : single linkage clustering A A B C HSP ≥ 80 % length Identity ≥ 35 % A B C Cluster A, B, C

Si. Li. X : single linkage clustering with alignment coverage constraints (Mièle et al. BMC Bioinformatics 2011) • Computing efficiency: – Ultra-fast – Memory efficient – Scalable (parallel architecture) • Clustering quality: – At least as good as the previously published methods

However … • Because of over-extension of BLAST alignments, some sequences that share only partial homology may be clustered in a same family • The risk of alignment over-extension is low, but becomes a problem for very large protein families • Use more stringent clustering criteria ? No : optimal clustering criteria are not the same for all families

Hi. Fi. X • The mode and tempo of evolution is specific to each protein family • A multiple alignment provides information about the specific pattern of evolution of a family • => this can be used to decide whether or not a new sequence belongs to that family

Hi. Fi. X • Step 1: rapid clustering (Si. Li. X) pre-families • Step 2: sub-clustering of pre-families into homogeneous protein clusters ⇒ sub-families • Step 3: progressive merging of sub-families into families, with evaluation of multiple alignment quality at each step ⇒ families

Hi. Fi. X

Results of clustering About 7, 000 proteins clustered into 300, 000 families Family size distribution: Number Sequences – at least 2 – 2: 10 – 10: 500 – 500: 2000 – more than 2000 Number of Families 296, 920 242, 398 53, 450 1, 026 79

Building Compare all proteins against each other (BLAST) Cluster homologous sequences into families (SILIX + HIFIX) Compute multiple alignments for each family Compute phylogenetic trees for each family Annotate phylogenetic trees (gene duplications, losses, transfers)

Compute multiple alignments All alignments ( ~ 300, 000) have been calculated with ClustalΩ

Building Compare all proteins against each other (BLAST) Cluster homologous sequences into families (SILIX + HIFIX) Compute multiple alignments for each family Compute phylogenetic trees for each family Annotate phylogenetic trees (gene duplications, losses, transfers)

Compute phylogenetic tree Question: what about the alternative splicing ?

Alternative splicing In eukaryotes, due to alternative splicing , one unique gene may be be transcripted into several transcripts

Transcripts in HOGENOM 6 We selected all the transcripts for each gene. Because the longest transcript is not allways the best!

Selection of a representaitive isoform in HOGENOM Because: We don’t want several proteins for a same gene in a phylogenetic tree: may be seen as a duplication We want 1 protein per gene for statistic comparison among organisms

Selection of a representaitive isoform : how ?

Selection of a representative isoform : how ? Eukarya 1 or more transcripts per gene Archaea and bacteria 1 transcript per gene

Selection of a representative isoform : how ? Eukarya clustering Archaea and bacteria

Selection of a representative isoform : how ? First step: when a gene has isoforms in different families ( ), choose a family for the gene

Selection of a representative isoform : how ? We select the family with the highest number of eukaryotic genes (and not proteins) 1 1 1 2 2 2 3 2 genes 3 genes

Selection of a representative isoform : how ? We select the family with the highest number of eukaryotic genes (and not proteins) If the number of eukaryotic genes are identical, we select the family with the highest number of eukaryotic proteins 1 1 1 2 2 2 3 2 genes 3 genes

Selection of a representative isoform : how ? We select the family with the highest number of eukaryotic genes (and not proteins) If the number of eukaryotic genes are identical, we select the family with the highest number of eukaryotic proteins 1 1 1 2 2 2 3 If the number of eukaryotic proteins are identical, we select the family with the highest number of proteins 2 genes 3 genes

Selection of a representative isoform : how ? We select the family with the highest number of eukaryotic genes (and not proteins) If the number of eukaryotic genes are identical, we select the family with the highest number of eukaryotic proteins 1 1 1 2 2 2 3 If the number of eukaryotic proteins are identical, we select the family with the highest number of proteins 3 genes The « rejected » isoforms are called « ISOFORMEX » SOME FAMILIES MAY FINALLY BE EMPTY AFTER THIS 2 genes

Selection of a representative isoform : how ? Second step: when a gene has isoforms in a family, choose a representative isoform for the gene 1 1 1 2 2 2 3 2 genes 3 genes

Selection of a representative isoform : how ? Second step: when a gene has isoforms in a family, choose a representative isoform for the gene 1 1 2 2 2 3 2 genes ? ? 1 2 genes 3 genes

Selection of a representative isoform : how ? We use the alignment

Selection of a representative isoform : how ? We use the alignment Suppression of ISOFORMEX

Selection of a representative isoform : how ? We use the alignment Selection positions with < 50% gap

Selection of a representative isoform : how ? 1 For each isoform of a given gene, for each position, we count for 1 each time the residue is identical to the residue in at least one of the isoforms of all other eukaryotic genes. 2 2 2 The isoform with the highest total is selected, the other isoforms being tagged as ISOFORMIN

Selection of a representative isoform : how ? 1 2 2 For each isoform of a given gene, for each position, we count for 1 each time the residue is identical to the residue in at least one of the isoforms of all other eukaryotic genes. The isoform with the highest total is selected, the other isoforms being tagged as ISOFORMIN

Selection of a representative isoform : how ? 1 2 2 2 2 For each isoform of a given gene, for each position, we count for 1 each time the residue is identical to the residue in at least one of the isoforms of all other eukaryotic genes. The isoform with the highest total is selected, the other isoforms being tagged as ISOFORMIN

Tree calculation

Tree calculation isformin a b c isformin d isformex e f g

Tree calculation Gblocks Phyml, Fast. Tree d isformin a b c d isformex e f g a b e c f g

Building Compare all proteins against each other (BLAST) Cluster homologous sequences into families (SILIX + HIFIX) Compute multiple alignments for each family Compute phylogenetic trees for each family Annotate phylogenetic trees (gene duplications, losses, transfers)

Annotate phylogenetic trees • Several methods are currently developed in the ANCESTROM project – Speciation, Duplication and Loss – Speciation, Duplication, Transfert and Loss – See Vincent Daubin talk tomorow