Many of the images in this powerpoint presentationare from Bioinformatics and Functional Genomicsby J Pevsner (ISBN 0-471-21004-8). Copyright © 2003 by Wiley.
These images and materials may not be usedwithout permission from the publisher.
Visit http://www.bioinfbook.org
Copyright notice
Multiple sequence alignment: today’s goals
• to define what a multiple sequence alignment is and how it is generated; to describe profile HMMs
• to introduce databases of multiple sequence alignments
• to introduce ways you can make your own multiple sequence alignments
• to show how a multiple sequence alignment provides the basis for phylogenetic trees
Page 319
Multiple sequence alignment: definition
Wallace, Blackshields, Higgins Curr Opp Struct Biol 2005 15:261-266
Multiple sequence alignment: properties
• for two proteins sharing 30% amino acid identity, about 50% of the individual amino acids are superposable in the two structures
Page 320
Generally align proteins:
nucleotides less well-conserved
nucleotide sequences are less informative -> harder to align with high confidence
How do you know if you have the “correct” alignment of a protein family? Is there one correct alignment?
Sequence identity (%)
Prop
ortio
n of
resi
dues
in c
omm
on c
ore
Proportion of structurally superposable residues in pairwise alignments
as a function of sequence identity
After Chothia & Lesk (1986)
GlobinCytochrome cSerine proteaseImmunoglobulin domain
0.5
0.25
0.75
100 75 50 25 0
Multiple sequence alignment: features
• some aligned residues, such as cysteines that formdisulfide bridges, may be highly conserved
• there may be conserved motifs such as a transmembrane domain
• there may be conserved secondary structure features
• there may be regions with consistent patterns ofinsertions or deletions (indels), often indicating the presence of subfamilies
Page 320
Multiple sequence alignment: uses
• MSA is more sensitive than pairwise alignmentin detecting homologs
• Detect conserved motifs and domains (from databasesearch or from alignment of a family)
• Phylogenetic analysis: MSA provides accurate estimationof evolutionary distances
• 2° and 3° structure prediction: MSA predictions significantlybetter than from a single sequence
• very few bioinformatics protocols bypass MSA stagePage 321
Multiple sequence alignment: methods
There are four main ways to make a multiple sequence alignment:
(1) Exact methods
(2) Progressive alignment (CLUSTALW, MUSCLE)
(3) Iterative approaches (Praline, IterAlign)
(4) Consistency-based methods (ProbCons)
Multiple sequence alignment: methods
Exact methods: dynamic programmingInstead of the 2-D dp matrix that you saw in theNeedleman-Wunsch technique, think about a 3-D,4-D etc matrix.
Exact methods give optimal alignments but are not feasible in time or space for more than ~10 sequences
Still an extremely active field though—the “holy grail”
Multiple sequence alignment: methods
Progressive methods: use a guide tree (a little like aphylogenetic tree but NOT a phylogenetic tree) to determine how to combine pairwise alignments one by oneto create a multiple alignment.
Making multiple alignments using trees was a verypopular subject in the ‘80s. Fitch and Yasunobu (1974)may have first proposed the idea, but Hogeweg andHesper (1984) and many others worked on the topic beforeFeng and Doolittle (1987) hit the scene—they made one important contribution that got their names attached to thisalignment method.
Examples: CLUSTALW, MUSCLE
Multiple sequence alignment: methods
Iterative methods: compute a sub-optimal solution and keep modifying that intelligently using dynamic programming or other methods until the solution converges.
Examples: IterAlign, Praline, MAFFT
Multiple sequence alignment: methods
Consistency-based algorithms: generally use a database of both local high-scoring alignments and long-range global alignments to create a final alignment
These are very powerful, very fast, and very accurate methods
Examples: T-COFFEE, Prrp, DiAlign, ProbCons
Multiple sequence alignment: methodsAaack! This is confusing! How do we know whichprogram to use?
There are benchmarking multiple alignment datasets that have been aligned painstakingly by hand, by structural similarity, or by extremely time- and memory-intensive automated exact algorithms.
Also, to be practical, some programs have interfaces that are much more user-friendly than others. And most programs are excellent so it depends on your preference.
If your proteins have 3D structures, USE THESE to help you judge your alignments.
Multiple sequence alignment: methods
Benchmarking tests suggest that ProbCons, a consistency-based/progressive algorithm, performs the best on the BAliBASE set, although MUSCLE, a progressive alignment package, is an extremely fast and accurate program.
CLUSTALW, everyone’s old favorite, continues to be a standout and is included in almost every MSA paper you will see. It has withstood the test of time. Plus, it has a nice interface (expecially with CLUSTALX) and is easy to use.
Multiple sequence alignment: methods
Example of progressive MSA using CLUSTALW: two data sets
Five distantly related lipocalins (human to E. coli)
Five closely related RBPs
When you do this, obtain the sequences of interest in FASTA format!
Page 321
Get sequences from GenBank
Get sequences from GenBank
Use Clustal W to do a progressive MSA
http://www.ebi.ac.uk/clustalw/
Feng-Doolittle MSA occurs in 3 stages
[1] Do a set of global pairwise alignments(Needleman and Wunsch)
[2] Create a guide tree
[3] Progressively align the sequences according to theguide tree
Page 321
Progressive MSA stage 1 of 3:generate global pairwise alignments
five distantly related lipocalins
best score
Progressive MSA stage 1 of 3:generate global pairwise alignments
five closely related lipocalins
best score
Number of pairwise alignments needed
Need to align every sequence to everyother sequence (handshake problem)
For N sequences, (N-1)(N)/2
For 5 sequences, (4)(5)/2 = 10
Page 322
Feng-Doolittle stage 2: guide tree
• Convert similarity scores to distance scores (seetext)
• A guide tree shows the distance between objects
• Use UPGMA (defined in Chapter 11)
• ClustalW provides a syntax to describe the tree
• A guide tree is not a phylogenetic tree
Page 323
Progressive MSA stage 2 of 3:generate a guide tree calculated from
the distance matrix
five distantly related lipocalins
Progressive MSA stage 2 of 3:generate guide tree
five closely related lipocalins
Feng-Doolittle stage 3: progressive alignment
• Make a MSA based on the order in the guide tree
• Start with the two most closely related sequences
• Then add the next closest sequence
• Continue until all sequences are added to the MSA
• Rule: “once a gap, always a gap.” <- this is what madethem famous!
Page 324
Page 324
Why “once a gap, always a gap”?
• There are many possible ways to make a MSA
• Where gaps are added is a critical question
• Gaps are first added to the first two (closest) sequences
• To change the initial gap choices later on would beto give more weight to distantly related sequences
• To maintain the initial gap choices is to trustthat those gaps are most believable (shouldn’t letthe final alignment be affected most by distantly related sequences!)
Page 324
Progressive MSA stage 3 of 3:progressively align the sequences
following the branch order of the tree
Fig. 10.3Page 324
Progressive MSA stage 3 of 3:CLUSTALX output
Clustal W alignment of 5 closely related lipocalins
CLUSTAL W (1.82) multiple sequence alignment
gi|89271|pir||A39486 MEWVWALVLLAALGSAQAERDCRVSSFRVKENFDKARFSGTWYAMAKKDP 50gi|132403|sp|P18902|RETB_BOVIN ------------------ERDCRVSSFRVKENFDKARFAGTWYAMAKKDP 32gi|5803139|ref|NP_006735.1| MKWVWALLLLAAW--AAAERDCRVSSFRVKENFDKARFSGTWYAMAKKDP 48gi|6174963|sp|Q00724|RETB_MOUS MEWVWALVLLAALGGGSAERDCRVSSFRVKENFDKARFSGLWYAIAKKDP 50gi|132407|sp|P04916|RETB_RAT MEWVWALVLLAALGGGSAERDCRVSSFRVKENFDKARFSGLWYAIAKKDP 50
********************:* ***:*****
gi|89271|pir||A39486 EGLFLQDNIVAEFSVDENGHMSATAKGRVRLLNNWDVCADMVGTFTDTED 100gi|132403|sp|P18902|RETB_BOVIN EGLFLQDNIVAEFSVDENGHMSATAKGRVRLLNNWDVCADMVGTFTDTED 82gi|5803139|ref|NP_006735.1| EGLFLQDNIVAEFSVDETGQMSATAKGRVRLLNNWDVCADMVGTFTDTED 98gi|6174963|sp|Q00724|RETB_MOUS EGLFLQDNIIAEFSVDEKGHMSATAKGRVRLLSNWEVCADMVGTFTDTED 100gi|132407|sp|P04916|RETB_RAT EGLFLQDNIIAEFSVDEKGHMSATAKGRVRLLSNWEVCADMVGTFTDTED 100
*********:*******.*:************.**:**************
gi|89271|pir||A39486 PAKFKMKYWGVASFLQKGNDDHWIIDTDYDTYAAQYSCRLQNLDGTCADS 150gi|132403|sp|P18902|RETB_BOVIN PAKFKMKYWGVASFLQKGNDDHWIIDTDYETFAVQYSCRLLNLDGTCADS 132gi|5803139|ref|NP_006735.1| PAKFKMKYWGVASFLQKGNDDHWIVDTDYDTYAVQYSCRLLNLDGTCADS 148gi|6174963|sp|Q00724|RETB_MOUS PAKFKMKYWGVASFLQRGNDDHWIIDTDYDTFALQYSCRLQNLDGTCADS 150gi|132407|sp|P04916|RETB_RAT PAKFKMKYWGVASFLQRGNDDHWIIDTDYDTFALQYSCRLQNLDGTCADS 150
****************:*******:****:*:* ****** *********
Fig. 10.5Page 326
Progressive MSA stage 3 of 3:progressively align the sequences
following the branch order of the tree:Order matters
THE LAST FAT CAT THE FAST CAT THE VERY FAST CAT THE FAT CAT
THE LAST FAT CATTHE FAST CAT ---
THE LAST FA-T CATTHE FAST CA-T ---THE VERY FAST CAT THE LAST FA-T CAT
THE FAST CA-T ---THE VERY FAST CATTHE ---- FA-T CATAdapted from C. Notredame, Pharmacogenomics 2002
Progressive MSA stage 3 of 3:progressively align the sequences
following the branch order of the tree:Order matters
THE FAT CAT THE FAST CAT THE VERY FAST CAT THE LAST FAT CAT
THE FA-T CATTHE FAST CAT
THE ---- FA-T CATTHE ---- FAST CATTHE VERY FAST CAT THE ---- FA-T CAT
THE ---- FAST CATTHE VERY FAST CATTHE LAST FA-T CATAdapted from C. Notredame, Pharmacogenomics 2002
Progressive MSA stage 3 of 3:progressively align the sequences
following the branch order of the tree:CLUSTALW results
MUSCLE: next-generation progressive MSA
1) Build a draft progressive alignmentDetermine pairwise similarity through k-mer counting
(not by alignment)
Compute distance (triangular distance) matrix
Construct tree using UPGMA
Construct draft progressive alignment following tree
MUSCLE: next-generation progressive MSA
2) Improve the progressive alignmentCompute pairwise identity through current MSA
Construct new tree with Kimura distance measures
Compare new and old trees: if improved, repeat this step, if not improved, then we’re done
MUSCLE: next-generation progressive MSA
3) Refinement of the MSASplit tree in half by deleting one edge
Make profiles of each half of the tree
Re-align the profiles
Accept/reject the new alignment
MUSCLE output (formattedwith SeaView)
Praline output for thesame alignment: pure iterative
approach
ProbCons—consistency-based approach
Combines iterative and progressive approaches with a unique probabilistic model.
Uses Hidden Markov Models (more in a minute) to calculate probability matrices for matching residues, uses this to construct a guide tree
Progressive alignment hierarchically along guide tree
Post-processing and iterative refinement (a little like MUSCLE)
ProbCons—consistency-based approach
ProbCons output for thesame alignment: how consistency
iteration helps
Multiple sequence alignment to profile HMMs• in 90’s people began to see that aligning sequences to profiles gave much more information than pairwise alignment alone.
• Hidden Markov models (HMMs) are “states”that describe the probability of having aparticular amino acid residue at arrangedIn a column of a multiple sequence alignment
• HMMs are probabilistic models
• Like a hammer is more refined than a blast,an HMM gives more sensitive alignmentsthan traditional techniques such as progressive alignments
Page 325
Simple Markov Model
Rain = dog may not want to go outside
Sun = dog will probably go outside
R
S
0.15
0.85
0.2
0.8
Markov condition = no dependency on anything but nearest previous state(“memoryless”)
Simple Hidden Markov Model
Observation: YNNNYYNNNYN
(Y=goes out, N=doesn’t go out)
What is underlying reality (the hidden state chain)?
R
S
0.15
0.85
0.2
0.8
P(dog goes out in rain) = 0.1
P(dog goes out in sun) = 0.85
GTWYA (hs RBP)GLWYA (mus RBP)GRWYE (apoD)GTWYE (E Coli)GEWFS (MUP4)
An HMM is constructed from a MSA
Example: five lipocalins
Fig. 10.6Page 327
GTWYAGLWYAGRWYEGTWYEGEWFS
Prob. 1 2 3 4 5p(G) 1.0p(T) 0.4p(L) 0.2p(R) 0.2p(E) 0.2 0.4p(W) 1.0p(Y) 0.8p(F) 0.2p(A) 0.4p(S) 0.2
Fig. 10.6Page 327
GTWYAGLWYAGRWYEGTWYEGEWFS
Prob. 1 2 3 4 5p(G) 1.0p(T) 0.4p(L) 0.2p(R) 0.2p(E) 0.2 0.4p(W) 1.0p(Y) 0.8p(F) 0.2p(A) 0.4p(S) 0.2
P(GEWYE) = (1.0)(0.2)(1.0)(0.8)(0.4) = 0.064
log odds score = ln(1.0) + ln(0.2) + ln(1.0) + ln(0.8) + ln(0.4) = -2.75 Fig. 10.6Page 327
GTWYAGLWYAGRWYEGTWYEGEWFS
P(GEWYE) = (1.0)(0.2)(1.0)(0.8)(0.4) = 0.064
log odds score = ln(1.0) + ln(0.2) + ln(1.0) + ln(0.8) + ln(0.4) = -2.75
G:1.0T:0.4L:0.2R:0.2E:0.2
W:1.0Y:0.8F:0.2
E:0.4A:0.4S:0.2
Fig. 10.6Page 327
Structure of a hidden Markov model (HMM)
M
Iy
Ix
p1
p7
p6
p5
p3
p2
p4
Structure of a hidden Markov model (HMM):Trellis representation
Fig. 10.7Page 328
HMMER: build a hidden Markov model
Determining effective sequence number ... done. [4]Weighting sequences heuristically ... done.Constructing model architecture ... done.Converting counts to probabilities ... done.Setting model name, etc. ... done. [x]
Constructed a profile HMM (length 230)Average score: 411.45 bitsMinimum score: 353.73 bitsMaximum score: 460.63 bitsStd. deviation: 52.58 bits
Fig. 10.8Page 329
HMMER: calibrate a hidden Markov model
HMM file: lipocalins.hmmLength distribution mean: 325Length distribution s.d.: 200Number of samples: 5000random seed: 1034351005histogram(s) saved to: [not saved]POSIX threads: 2- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
HMM : xmu : -123.894508lambda : 0.179608max : -79.334000
Fig. 10.8Page 329
HMMER: search an HMM against GenBankScores for complete sequences (score includes all domains):Sequence Description Score E-value N-------- ----------- ----- ------- ---gi|20888903|ref|XP_129259.1| (XM_129259) ret 461.1 1.9e-133 1gi|132407|sp|P04916|RETB_RAT Plasma retinol- 458.0 1.7e-132 1gi|20548126|ref|XP_005907.5| (XM_005907) sim 454.9 1.4e-131 1gi|5803139|ref|NP_006735.1| (NM_006744) ret 454.6 1.7e-131 1gi|20141667|sp|P02753|RETB_HUMAN Plasma retinol- 451.1 1.9e-130 1..gi|16767588|ref|NP_463203.1| (NC_003197) out 318.2 1.9e-90 1
gi|5803139|ref|NP_006735.1|: domain 1 of 1, from 1 to 195: score 454.6, E = 1.7e-131*->mkwVMkLLLLaALagvfgaAErdAfsvgkCrvpsPPRGfrVkeNFDv
mkwV++LLLLaA + +aAErd Crv+s frVkeNFD+gi|5803139 1 MKWVWALLLLAA--W--AAAERD------CRVSS----FRVKENFDK 33
erylGtWYeIaKkDprFErGLllqdkItAeySleEhGsMsataeGrirVL+r++GtWY++aKkDp E GL+lqd+I+Ae+S++E+G+Msata+Gr+r+L
gi|5803139 34 ARFSGTWYAMAKKDP--E-GLFLQDNIVAEFSVDETGQMSATAKGRVRLL 80
eNkelcADkvGTvtqiEGeasevfLtadPaklklKyaGvaSflqpGfddy+N+++cAD+vGT+t++E dPak+k+Ky+GvaSflq+G+dd+
gi|5803139 81 NNWDVCADMVGTFTDTE----------DPAKFKMKYWGVASFLQKGNDDH 120
Fig. 10.9Page 330
HMMER: search an HMM against GenBankmatch to a bacterial lipocalin
gi|16767588|ref|NP_463203.1|: domain 1 of 1, from 1 to 177: score 318.2, E = 1.9e-90*->mkwVMkLLLLaALagvfgaAErdAfsvgkCrvpsPPRGfrVkeNFDv
M+LL+ +A a ++ Af+v++C++p+PP+G++V++NFD+gi|1676758 1 ----MRLLPVVA------AVTA-AFLVVACSSPTPPKGVTVVNNFDA 36
erylGtWYeIaKkDprFErGLllqdkItAeySleEhGsMsataeGrirVL+rylGtWYeIa+ D+rFErGL + +tA+ySl++ +G+i+V+
gi|1676758 37 KRYLGTWYEIARLDHRFERGL---EQVTATYSLRD--------DGGINVI 75
eNkelcADkvGTvtqiEGeasevfLtadPaklklKyaGvaSflqpGfddyNk++++D+ +++ +EG+a ++t+ P +++lK+ Sf++p++++y
gi|1676758 76 -NKGYNPDR-EMWQKTEGKA---YFTGSPNRAALKV----SFFGPFYGGY 116
Fig. 10.9Page 330
HMMER: search an HMM against GenBankScores for complete sequences (score includes all domains):Sequence Description Score E-value N-------- ----------- ----- ------- ---gi|3041715|sp|P27485|RETB_PIG Plasma retinol- 614.2 1.6e-179 1gi|89271|pir||A39486 plasma retinol- 613.9 1.9e-179 1gi|20888903|ref|XP_129259.1| (XM_129259) ret 608.8 6.8e-178 1gi|132407|sp|P04916|RETB_RAT Plasma retinol- 608.0 1.1e-177 1gi|20548126|ref|XP_005907.5| (XM_005907) sim 607.3 1.9e-177 1gi|20141667|sp|P02753|RETB_HUMAN Plasma retinol- 605.3 7.2e-177 1gi|5803139|ref|NP_006735.1| (NM_006744) ret 600.2 2.6e-175 1
gi|5803139|ref|NP_006735.1|: domain 1 of 1, from 1 to 199: score 600.2, E = 2.6e-175*->meWvWaLvLLaalGgasaERDCRvssFRvKEnFDKARFsGtWYAiAK
m+WvWaL+LLaa+ a+aERDCRvssFRvKEnFDKARFsGtWYA+AKgi|5803139 1 MKWVWALLLLAAW--AAAERDCRVSSFRVKENFDKARFSGTWYAMAK 45
KDPEGLFLqDnivAEFsvDEkGhmsAtAKGRvRLLnnWdvCADmvGtFtDKDPEGLFLqDnivAEFsvDE+G+msAtAKGRvRLLnnWdvCADmvGtFtD
gi|5803139 46 KDPEGLFLQDNIVAEFSVDETGQMSATAKGRVRLLNNWDVCADMVGTFTD 95
tEDPAKFKmKYWGvAsFLqkGnDDHWiiDtDYdtfAvqYsCRLlnLDGtCtEDPAKFKmKYWGvAsFLqkGnDDHWi+DtDYdt+AvqYsCRLlnLDGtC
gi|5803139 96 TEDPAKFKMKYWGVASFLQKGNDDHWIVDTDYDTYAVQYSCRLLNLDGTC 145
Fig. 10.9Page 330
BLOCKSCDDDOMOINTERPROiProClassMetaFAMPFAMPRINTSPRODOM PROSITESMART
Databases of multiple sequence alignments
Page 331-332
BLOCKS (HMM)CDD (HMM)DOMO (Gapped MSA)INTERPROiProClassMetaFAMPfam (profile HMM library)PRINTSPRODOM (PSI-BLAST)PROSITESMART
Databases of multiple sequence alignments
Page 331-332
BLOCKSCDDDOMOINTERPROiProClassMetaFAMPFAMPRINTSPRODOM PROSITESMART
Databases of multiple sequence alignments
Integrative resources
Page 331-332
Pfam (protein family) database:http://pfam.wustl.edu/ or
http://www.sanger.ac.uk/Software/Pfam/
Pfam (protein family) database
Pfam (protein family) database
Pfam (protein family) database
Pfam (protein family) database
SMART: Simple ModularArchitecture Research Tool(emphasis on cell signaling)Search for conserved domains
Page 338
BLOCKSCDDDOMOINTERPROiProClassMetaFAMPFAMPRINTSPRODOM PROSITESMART
Databases of multiple sequence alignments
CDD at NCBI = PFAM + SMART
Page 333
Query = your favorite protein
Database = set of many PSSMs
CDD is related to PSI-BLAST, but distinct
CDD searches against profiles generatedfrom pre-selected alignments
Purpose: to find conserved domainsin the query sequence
You can access CDD via DART at NCBI
CDD uses RPS-BLAST: reverse position-specific
Page 333
CDART (fromNCBI) allowsCDD searches
BLOCKS server http://blocks.fhcrc.org/
BLOCKS: ungapped MSA
Very highly conserved regions
Fig. 10.22Page 341
See chapter 8
PROSITE has a syntax to define signatures
See chapter 8
PopSet: polymorphisms in population data
Fig. 10.23Page 342
PopSet: polymorphisms in population data
Fig. 10.23Page 342
PopSet: polymorphisms in population data
http://bioweb.pasteur.fr/seqanal/alignment/ReviewEdital/review-edital-desc.html
Nice overview site—lots of alignment utilities, including editors and viewers. Most are current.
http://www.public.iastate.edu/~pedro/research_tools.html
Pedro’s molecular biology tools. Most are current.
Big Picture sites–very useful
Manual curation:PfamPROSITEBLOCKSPRINTS
Automated curation:DOMOPRODOMMetaFam
+ comprehensive- alignment errors
MSA databases: manual vs. automated curation
AMASCINEMAClustalWClustalXDIALIGNHMMTMatch-BoxMultAlinMSAMuscaPileUpSAGAT-COFFEE
Multiple sequence alignment programs
Fig. 10.25Page 346
Selecting sequences for a PileUp in GCG
GCGPileUp
Fig. 10.25Page 346
GCGPileUp
Fig. 10.26Page 347
Multiple sequence alignment algorithms
Progressive
Iterative
Local Global
PIMA
DIALIGN SAGA
CLUSTALPileUpother
Fig. 10.27Page 348
[1] Create or obtain a database of protein sequencesfor which the 3D structure is known. Thus we candefine “true” homologs using structural criteria.
[2] Try making multiple sequence alignmentswith many different sets of proteins (very related,very distant, few gaps, many gaps, insertions,outliers).
[3] Compare the answers.
Strategy for assessment of alternativemultiple sequence alignment algorithms
Page 346
Commonly used benchmarking databases:
BAliBASESMARTSABmarkPREFAB (Protein Reference Alignment Benchmark)
Strategy for assessment of alternativemultiple sequence alignment algorithms
[1] As percent identity among proteins drops,performance (accuracy) declines also. This isespecially severe for proteins < 25% identity.
Proteins <25% identity: 65% of residues align well
Proteins <40% identity: 80% of residues align well
Conclusions: assessment of alternativemultiple sequence alignment algorithms
Page 350
[2] “Orphan” sequences are highly divergent members of a family. Surprisingly, orphans do notdisrupt alignments.
Conclusions: assessment of alternativemultiple sequence alignment algorithms
Page 350
[3] Separate multiple sequence alignments can be combined (e.g. RBPs and lactoglobulins).
Profile-based methods are very powerful but the strength of the profile depends on the quality ofthe initial alignment.
Conclusions: assessment of alternativemultiple sequence alignment algorithms
Page 350
[4] When proteins have large N-terminal or C-terminal extensions, local alignment algorithmsare superior. PileUp (global) is an exception.
Conclusions: assessment of alternativemultiple sequence alignment algorithms
Page 350