Multiple Sequence Alignment MSA 1 Multiple Alignment Number

Multiple Sequence Alignment (MSA) 1

Multiple Alignment • Number of sequences >2 • Global alignment • Seek an alignment that maximizes score 2

Multiple Alignment Multiple sequence alignment can be viewed as an extension of pairwise sequence alignment, but the complexity of the computation grows exponentially with the number of sequences being considered and their lengths and, therefore, it is not feasible to search exhaustively for the optimal alignment even for even a modest number of short sequences. For example, there approximately 1038 possible alignments that can be produced from five 10 -nucleotide-long sequences 3

Possible Alignments TCGGCTCGAC GAGCGTGATGGCGAGC--CGTCGTAAC- TCG-GC-TCGAC GAGCGT-GAT-G-GCG-AG---C CG-TCGTA--AC … 1038 4

Multiple Alignment • In addition to the scores employed by pairwise alignment, we have additional scores. • However, we still seek an alignment that maximizes score 5

Similarity Scoring Scheme {4, 0, 0} = 1 {3, 1, 0, 0, 0} = 0. 75 {2, 2, 0, 0, 0} = 0. 5 {2, 1, 1, 0, 0} = 0. 5 {1, 1, 0} = 0 5 character states: A, C, T, G, – 6

Two Possible Alignments GTCGTAGTCGGCTCGAC GTCTAGCGTGATGCGAAGAGGCGAGC--GCCGTCGTAAC- 1*1 + 3*0. 75 + 12*0. 5 = 9. 25 4*1 + 13*0. 75 + 2*0. 5 = 11. 75 GTCGTAGTCG-GC-TCGAC GTC-TAG-CGAGCGT-GAT GC-GAAG-AG-GCG-AG-C GCCGTCG-CG-TCGTA-AC 7

Alignments can be easy or difficult Easy Difficult 8

Multiple Alignment • Dynamic programming (exhaustive, exact) – Consider 2 protein sequences of 100 amino acids in length. – If it takes 1002 (103) seconds to exhaustively align these sequences, then it will take 104 seconds to align 3 sequences, 105 to align 4 sequences, etc. – It will take ~1021 seconds to align 20 sequences. One year is ~3 ✕ 107 seconds. The age of the visible universe is ~1018 seconds. • Progressive alignment (heuristic, approximate) 9

Progressive Alignment • Devised by Feng and Doolittle in 1987. • Essentially a heuristic method and, as such, not guaranteed to find the “optimal” or “best” alignment. • Requires pairwise alignments as a starting point • One of the first successful implementation was Clustal (by Des Higgins et al. ) 10

• Sum of pairwise alignment scores – For n sequences, there are n(n-1)/2 pairs • Based on pairwise alignment scores – Build n by n table of pairwise scores • Align similar sequences first – After alignment, consider as single sequence – Continue aligning with further sequences GTCGTAGTCG-GC-TCGAC GTC-TAG-CGAGCGT-GAT GC-GAAG-AG-GCG-AG-C GCCGTCG-CG-TCGTA-AC 12

First step: A B C D Compute the pairwise alignments for all against all (6 pairwise alignments) the similarities are stored in a table A B C D A B 11 C 3 1 D 2 2 10 13

Second step: A B C D A B 11 Cluster the sequences to create a tree (guide tree): • Represents the order in which pairs of sequences are to be aligned • Similar sequences are neighbors in the tree • Distant sequences are distant from each other in the tree Guide tree C 3 1 D 2 2 10 A B C D 14

Guide Tree A guide tree is not a phylogenetic tree! 15

Third step: Align most similar pairs A B C D Align the alignments as if each of them was a single sequence (replace with a single consensus sequence or use a profile) 16

Alignment of alignments M Q T F L H T W L Q S W X Y M L L L M Q H Q - T T S T T I I F W W F W L T I F M T I W 17

Input File > Usually FASTA format 19

Input file > Sulfolobus acidocaldarius gi|152927|gp|J 03218|SSOATPMA_1 MVSEGRVVRVNGPLVIADGMREAQMFEVVYVSDLKLVGEITRIE > Thermococcus sp. gi|2605627 ATPase alpha subunit MGRIIRVTGPLVVADGMKGAKMYEVVRVGEMGLIGEIIRLEGDKAVIQVYEETAGIRPG EPVEGTGSSLS > Acetabularia acetabulum gi|1303673|gnl|PID|d 1009732 adenosine triphosphatase A subunit MSKAKEGDYGSIKKVSGPVVVADNMGGSAMYELVRVGTGELIGEIIRLEGDTATIQVYE ETSGLTVGDGV … Output file 20

MSA: Problems Sequences that are similar only in some smaller regions may be misaligned because MSA tries to find global, rather than local alignments. Sequence that contains a large insertion compared to the rest may be misaligned because MSA tries to find global, rather than local alignments. 21

MSA: Problems Sequence that contains a repetitive element (such as a domain), while another sequence only contains one copy. VS 22

MSA: Problems • Pairwise alignment is an optimal algorithm. • Multiple alignment is not an optimal algorithm. Better alignments might exist! • The algorithm yields a possible alignment, but not necessarily the best one. 23

MSA: Problems • Because of the progressive methodology, errors at the beginning of the alignment are more important than errors at the end of the alignment. 24

Clustal: Advice • Sequence weighting (Should we treat all the sequences as equals? ) • Position weighting (Should we treat all positions in the sequences as equals? ) • Varying substitution matrices • Residue-specific gap penalties and reduced penalties in hydrophilic regions (external regions of protein sequences), encourage gaps in loops rather than in core regions. • Positions in early alignments where gaps have been opened receive locally reduced gap penalties to encourage openings in subsequent alignments • Vary “gap opening penalty” and “gap extension penalty” • Discourage too many gaps, too close to one another • Take into account hydrophilic and hydrophobic structures • Avoid divergent sequences, as they are the most difficult to align 25

Alignment of protein-coding DNA sequences • It is not very sensible to align the DNA sequences of protein-coding genes. ATGCTGTTAGGG ATGCTCGTAGGG ATGCT-GTTAGGG ATGCTCGTA-GGG The result might be implausible and might not reflect what is known about biological processes. It is much more sensible to translate the sequences into their corresponding amino acid sequences, align these protein sequences and then put the gaps in the DNA sequences according to the amino acid alignment. 26

Effect of gap penalties on amino-acid alignment Human pancreatic hormone precursor versus chicken pancreatic hormone (a) Penalty for gaps is 0 (b) Penalty for a gap of size k nucleotides is wk = 1 + 0. 1 k (c) The same alignment as in (b), only the similarity between the two sequences is further enhanced by showing pairs of 27 biochemically similar amino acids

Anchored alignment: The anchored-alignment procedure uses expert knowledge to improve multiple alignments. The user can specify a list of anchor points, points each of which consists of a pair of equallength segments that are to be aligned by the program. If residue x from one of the input sequences is paired to residue y from another sequence, then y is the only residue that can be aligned to x, and vice versa. All residues to the left and the right of x are aligned, respectively, to the residues to the left and the right of y. 28

Consider, for example, the following example: >seq 1 WKKNADAPKRAMTSFMKAAY >seq 2 WNLDTNSPEEKQAYIQLAKDDRIRYD >seq 3 WRMDSNQKNPDSNNPKAAYNKGDANAPK The non-anchored default version of DIALIGN would calculate the following alignment for this input sequence set: seq 1 1 WKKNAD-----APKRamtsfm. KAAY------seq 2 1 WNLDTN-----SPEE------KQAYiqla. KDDriryd seq 3 1 WRMDSNqknpd. SNNP------KAAYn---KGDsnapk 29

Now let's assume, the user has some expert knowledge about a certain domain that is present in all of the input sequences; the domains in the three sequences are thought to be homologous to each other: >seq 1 WKKNADAPKRAMTSFMKAAY >seq 2 WNLDTNSPEEKQAYIQLAKDDRIRYD >seq 3 WRMDSNQKNPDSNNPKAAYNKGDANAPK The user wants to define this motif as anchor and align the rest of the sequences automatically, given the pre-defined constraints imposed by this anchor. Since anchor points are defined as pairs of equal-length segments, we need two anchor points to enforce alignment of the above motif. 30

For example, one could choose Anchor point 1: >seq 1 WKKNADAPKRAMTSFMKAAY >seq 2 WNLDTNSPEEKQAYIQLAKDDRIRYD >seq 3 WRMDSNQKNPDSNNPKAAYNKGDANAPK Anchor point 2: >seq 1 WKKNADAPKRAMTSFMKAAY >seq 2 WNLDTNSPEEKQAYIQLAKDDRIRYD >seq 3 WRMDSNQKNPDSNNPKAAYNKGDANAPK If the above motif is to be aligned by our program, these two anchor points need to be specified. 31

Format for user-defined anchor points: To specify a set of anchor points, a file with the coordinates of these anchor points is needed. Since each anchor point corresponds to a equal-length segment pair involving two of the input sequences, coordinates for anchor points are defined as follows: (1) first sequence involved (2) second sequence involved (3) start of anchor in first sequence (4) start of anchor in second sequence (5) length of anchor. (6) specify a score of an anchor point. This score is necessary to prioritize anchor point in case they are inconsistent with each other, i. e. , if not all of them can be used simultaneously for the same alignment. the above coordinates (in the above given order). 32

1 2 4 6 6 Thus, the above two anchor points are specified as follows: 1 2 4 6 6 4. 5 2 3 6 11 8 1. 3 4. 5 and 1. 3 are (arbitrary) scores for anchor points 1 and 2. 33

MSA Approaches • Progressive approach: CLUSTALW (CLUSTALX) T-COFFEE PILEUP • Iterative approach (Repeatedly realigned subsets of sequences): Mult. Alin, Di. Align • Statistical Methods (e. g. , Hidden Markov Models) SAM 2 K • Genetic algorithms SAGA 34

T-coffee • An MSA program • Uses principles similar to Clustal • Combining sequence alignment method with structural alignment techniques • More accurate but longer running time • Limits to the number of sequences it can align (~100) 35

MSA: Problems Giddy Landan and Dan Graur. 2007. Heads or Tails: A Simple Reliability Check for Multiple Sequence Alignments. Molecular Biology and Evolution 24(6): 1380 -1383. 36

Genomic alignment (with MAUVE)