7 Multiple Genome Alignment

Transcription

1 94 Bioinformatics I, WS /3, D. Huson, December 3, 0 7 Multiple Genome Alignment Assume we have a set of genomes G,..., G t that we want to align with each other. If they are short and very closely related, then one can use a multiple sequence alignment tool such as ClustalW to align them. Example: Generally, genomes will differ by: G A C C A A T T A C C T T - - A T C C C T A C - - T A A A C T subsitutions, insertions or deletions of bases, but also: replacements, insertions, deletions, duplications, translocations or inversions of whole segments of the sequence. If differences of the latter type occur, then simple multiple sequence alignment is not sufficient and a method is required that explicitly takes such large-scale rearrangments into account. The typical situation looks like this: Genome A B C D Genome A' C' B' D' Genome 3 C'' B'' E D'' 7. Multiple Alignment of Rearranged Sequences In the comparison of multiple genomes, we must solve the following computational problem: Multiple Alignment of Rearranged Genomic Sequences Problem: Input: Genome sequences G,..., G t. Output: Set of conserved segments and alignments between them. 7. Mauve Mauve is a widely used tool that addresses the stated problem. Here is an overview: A. Darling et al., Genome Research (004)

2 Bioinformatics I, WS /3, D. Huson, December 3, 0 95 (a) Quickly find long stretches of sequences that match two or more genomes, these are called anchors. Example: anchor G (b) Form a set of locally colinear blocks (LCBs) from the anchors. G LCB LCB LCB 3 3 LCB (c) Recursively try to improve LCBs. (d) Align the set of sequences in each LCB. 7.. MUMs and Multi MUMs Consider two genome sequences G and. Recall that a MUM (maximal unique match) is a subsequence that occurs exactly once in both G and and cannot be extended. A MUM can be written as M = (L, p, p ), where L is the length and p i is the left most position in G i. If the match is in the reverse complement of G i, then we use p i to denote its position in G i. MUMs can be computed using a suffix tree. Consider genomes G,..., G t. A multi-mum is a subsequence present in t genomes, in each contained exactly once and not extentable. It is written as M = (l, p, p,..., p t ), where l = L(M) is the length of the subsequence and p i = p i (M) is it s position in the i-th genome. Set p i = 0 to indicate that i-th genome is not involved in M. The multiplicity of M is d(m) = {i p i 0}, the number of genomes involved in M. Mauve uses multi-mums as anchors. 7.. Computation of multi-mums Multi-MUMs can be computed using a suffix tree (assignment). However, Mauve uses a simple seed-and-extend approach to find all multi-mums of length k, as follows:

3 96 Bioinformatics I, WS /3, D. Huson, December 3, 0 (a) Generate a hash table that maps each k-mer to a list of its occurrences in genomes G,..., G t. (b) For each k-mer that maps to two or more different genomes, extend all corresponding matches to the left and right until the first mismatch is found: k-mer match A Extend T A T C A The runtime of this approach is reported as: O(t n + tn log tn), where t is the number of genomes and n is their average length Anchors in Mauve The set of anchors used by Mauve is initially the set m of all multi-matches of full multiplicity t Determining LCBs Mauve uses breakpoint analysis to partition M into a set of L of LCBs, where each LCB consists of a set of matches (multi-mums) M,..., M f with p i (M j ) p i (M j+ ) i =,..., t and j =,..., f. G M M M 3 M M M 3 or To improve the set of LCBs, repeat the following until all remaining LCBs have a given minimum weight w: Let B be an LCB of smallest weight. Remove all matches in B from M. Recompute the set of LCBs from M. Here, the weight of LCB is the number of nucleotides that its matches cover. Example: G 3 3

4 Bioinformatics I, WS /3, D. Huson, December 3, 0 97 If the weight of LCB- is w then remove all matches of LCB- from M and then recalculate the set of LCBs: G G G3 Second illustration of block removal: 7..5 Multiple sequence alignment Once the set L of LCBs has been calculated, the last step is to compute a MSA for each LCB. For each LCB, Mauve computes a progressive alignment in three steps:. Pairwise alignment of all sequences to get a distance matrix D.. Compute a guide tree from D. 3. Progressively build a MSA by pairwise alignment of profiles along the guide tree. Step () is the most time-consuming operation in any progressive alignment approach. To avoid repeately performing steps () and (), Mauve computes a simple guide tree from the set of multimums by applying Neighbor-Joining to the following distances: d(gi, Gj ) = # bases shared by Gi and Gj Gi + Gj, based on the amount of identical sequence shared. This tree is used as guide tree for all LCB alignments.

5 98 Bioinformatics I, WS /3, D. Huson, December 3, Bacterial Example The Mauve paper presents the analysis of nine enterobacterial genomes, namely four E.coli, two Shigella and three Salmonella genomes : Species Genome size Reference E. coli K MG655 4,639, Blattner et al. 997 E. coli O57:H7 EDL933 5,54,97 Perna et al. 00 E. coli O57:H7 VT- Saka 5,498,450 Hayashi et al. 00 E. coli CFT073 5,3,48 Welch et al. 00 S. flexneri A 457T 4,599,354 Wei et al. 003 S. flexneri A 4,607,03 Jin et al. 00 S. enterica Typhimurium LT 4,857,43 McClelland et al. 00 S. enterica Typhi CT8 4,809,037 Parkhill et al. 00 S. enterica Typhi Ty 4,79,96 Deng et al. 003 Processing by Mauve took 3 hours on a single core. This is the tree that Mauve estimated from the genome sequences and which is used to generate multiple alignments on each of the locally colinear blocks: Here is the resulting genome alignment: A. Darling et al., Genome Research (004)

6 Bioinformatics I, WS /3, D. Huson, December 3, 0 99 Question: How much sequence is conserved over all nine enterobacterial genomes? To address this, define a conserved backbone segment as a segment of a Mauve alignment that contains more than 50 gap-free columns and does not have a run of 50 or more consecutive gaps in any single genome sequence. Based on this, the nine enterobacteria have.86 Mb of conserved backbone sequence partitioned into 5 backbone segments. 7.4 Eukaryotes Example Although Mauve was originally designed for bacterial-sized genomes, it can also be applied to large genomes as well. The calculation of a comparison of human, mouse and rat (without calculating the alignments of reach of the LCB s) takes about hours. Depending on the parameters used, this alignment has between 000 and 000 LCBs:

7 00 Bioinformatics I, WS /3, D. Huson, December 3, Yersinia Example In 00, only three species of the enterobacterial genus Yersinia, ones that cause invasive human diseases (Yersinia pestis, Yersinia pseudotuberculosis, and Yersinia enterocolitica), had been sequenced. However, there were no genomic data on the Yersinia species with more limited virulence potential, such as are frequently found in soil and water environments. A paper published in 003 provided new genome sequences for other Yersinia species. The paper presents a number of different types of comparisons of the different genomes, including a Mauve analysis: Image source: P. Chen et al, Genome Biology (00) 3 P. Chen et al, Genome Biology (00),