Transcript O - Biotec
Sequence comparison
and Phylogeny
based on
Chapter 4
Lesk, Introduction to Bioinformatics
Michael Schroeder
BioTechnological Center
TU Dresden
Biotec
Contents
Motivation
Sequence comparison and alignments
Dot plots
Dynamic programming
Substitution matrices
Dynamic programming: Local and global alignments
and gaps
BLAST
Significance of alignments
Multiple sequence alignments
Phylogenetic trees
By Michael Schroeder, Biotec, 2004
2
Motivation
From where are we?
Recent Africa vs. Multi-regional Hypothese
In 1999 Encephalitis caused by the West Nile Virus
broke out in New York. How did the virus come to
New York?
How did the nucleus get into the eucaryotic cells?
To answer such questions we will need sequence
comparison and phylogenetic trees
By Michael Schroeder, Biotec, 2004
3
Sequence Similarity Searches
Sequence similarity can be clue to common evolutionary
ancestor…
E.g. globin genes in chimpanzees and humans
… or common function
E.g. v-sys onco genes in simian sarcoma virus leading to cancer
in monkeys and the seemingly unrelated growth stimulating
hormone PDGF, which stimulates cell growth (first success of
similarity idea, 1983)
In general:
If an unknown sequences is found, deduce its
function/structure indirectly by finding similar sequences, whose
function/structure is known
Assumption: Evolution changes sequences “slowly” often
maintaining main features of a sequence’s function/structure
By Michael Schroeder, Biotec, 2004
4
Sequence alignment
Substitutions, insertions and deletions can be
interpreted in evolutionary terms
But: distinguish chance similarity and real biological
relationship
CCGTAA
TCGTAA
TCGTAC
TTGTAA
TAGTAC
By Michael Schroeder, Biotec, 2004
CCGTAT
TCGTAG
5
Evolution
Convergent evolution: same sequence evolved from
different ancestors
Back evolution - mutate to a previous sequence
CCGTAA
TCGTAA
TAGTAA
TAGTAC
TCGTAC
TAGTAC
By Michael Schroeder, Biotec, 2004
CCGTAT
TCGTAG
CCGTAA
6
Similarity vs. Homology
Any sequence can be similar
Sequences homologues if evolved from common
ancestor
Homologous sequences:
Orthologs: similar biological function
Paralogs: different biological function (after gene
duplication), e.g. lysozyme and α-lactalbumin, a
mammalian regulatory protein
Assumption: Similarity indicator for homology
Note, altered function of the expressed protein will
determine if the organism will survive to reproduce, and
hence pass on the altered gene
By Michael Schroeder, Biotec, 2004
7
Sequence alignments
Given two or more sequences, we wish to
Measure their similarity
Determine the residue-residue correspondences
Observe patterns of conservation and variability
Infer evolutionary relationships
By Michael Schroeder, Biotec, 2004
8
What is the best alignment?
Uninformative:
-------gctgaacg
ctataatc------ Without gaps:
gctgaacg
ctataatc
With gaps:
gctga-a--cg
--ct-ataatc
Another one:
gctg-aa-cg
-ctataatc Formally: The best alignment have only a minimal
number of mismatches (insertions, deletions,
replace)
We need a method to systematically explore and to
compute alignments
By Michael Schroeder, Biotec, 2004
9
Scores for an alignment
Percentage of matches
Score each match, mismatch, gap opening, gap extension
Example
match +1
mismatch -1
Gap opening -3
Gap extension -1
Uninformative: 0%, score= -21
Without gaps:25%,score= -4
With gaps: 0%, score= -23
Another one: 50%, score=-12
By Michael Schroeder, Biotec, 2004
-------gctgaacg
ctataatc------gctgaacg
ctataatc
gctga-a--cg
--ct-ataatc
gctg-aa-cg
-ctataatc-
10
Scores for an alignment
Percentage of matches
Score each match, mismatch, gap opening, gap extension
Example
match +2
mismatch -1
Gap opening -1
Gap extension -1
Uninformative: 0%, score= -17
Without gaps:25%,score= -2
With gaps: 0%, score= -15
Another one: 50%, score=0
By Michael Schroeder, Biotec, 2004
-------gctgaacg
ctataatc------gctgaacg
ctataatc
gctga-a--cg
--ct-ataatc
gctg-aa-cg
-ctataatc-
11
Dot plots
By Michael Schroeder, Biotec, 2004
12
Dot plots
A convenient way of comparing 2 sequences visually
Use matrix, put 1 sequence on X-axis, 1 on Y-axis
Cells with
identical characters filled with a ‘1’,
non-identical with ‘0’
(simplest scheme - could have weights)
By Michael Schroeder, Biotec, 2004
13
Dot plots
D O R O T H Y C R O W F O O T H O D G K I
N
D
O
R
O
T
H
Y
H
O
D
G
K
I
N
By Michael Schroeder, Biotec, 2004
14
Dot plots
D O R O T H Y C R O W F O O T H O D G K I
D D
O
O
O
R
R
O
O
O
T
T
H
H
Y
Y
H
H
O
O
O
D D
G
K
I
N
By Michael Schroeder, Biotec, 2004
N
D
O
O O
O
O
O O
O
R
T
H
H
O
O O
O
D
G
K
I
N
15
Interpreting dot plots
What do identical sequences look like?
What do unrelated sequences look like?
What do distantly related sequences look like?
What does reverse sequence look like?
Relevant for detections of stems in RNA structure
What does a palindrome look like?
Relevant for restriction enzymes
What do repeats look like?
What does a protein with domains A and B and another
one with domains B and C look like?
By Michael Schroeder, Biotec, 2004
16
Dot plot for identical sequences
D
D
O
R
O
T
H
H
O
D
O
D
G
K
I
N
D
O
R
O
O
O
O
R
O
O
T
T
H
H
Y
H
Y
H
H
O
D
Y
O
D
G
K
I
N
By Michael Schroeder, Biotec, 2004
O
H
O
D
G
K
I
N
17
Dotplot for unrelated sequences
D O R O T H Y H O D G K I
O
O
O
O
T
T
T
T
O
O
N
O
O
D D
D
I
I
E
T
T
E
R
R
By Michael Schroeder, Biotec, 2004
18
Dotplot for distantly related sequences
D
O
R
O
T
T
H
Y
H
O
D
G
K
I
N
T
I
I
M
O
O
O
T
O
T
H
H
Y
H
Y
J
E
N
N
K
K
I
I
N
N
By Michael Schroeder, Biotec, 2004
19
Dotplot for reverse sequences
D
O
R
O
T
H
Y
H
O
D
G
K
I
N
N
I
I
K
K
G
D
G
D
O
D
O
O
O
H
H
Y
H
Y
H
H
T
H
T
O
O
R
O
O
O
O
R
O
D
N
O
D
By Michael Schroeder, Biotec, 2004
20
Dotplot for reverse sequences
Relevant to identify stems in RNA structures
Plot sequence against its reverse complement
By Michael Schroeder, Biotec, 2004
21
Palindromes and restriction enzymes
Madam, I'm Adam
Able was I ere I saw Elba (supposedly said by Napoleon)
Doc note I dissent, a fast never prevents a fatness, I diet on cod.
Because DNA is double stranded and the strands run
antiparallel, palindromes are defined as any double stranded
DNA in which reading 5’ to 3’ both are the same
The HindIII cutting site:
– 5'-AAGCTT-3'
– 3'-TTCGAA-5'
The EcoRI cutting site:
– 5'-GAATTC-3'
– 3'-CTTAAG-5'
By Michael Schroeder, Biotec, 2004
22
Dotplot of a Palindrome
M
M
A
A
M
A
M
M
A
D
A
D
A
M
D
A
M
By Michael Schroeder, Biotec, 2004
A
M
23
Dotplot of repeats
T
T
W
E
T
W
T
T
W
E
N
T
Y
T
E
N
T
T
O
T
E
N
T
T
T
Y
T
W
W
N
T
Y
T
T
W
T
T
W
O
W
O
T
W
O
T
T
W
E
T
T
W
E
N
E
N
T
W
E
N
N
T
Y
T
T
Y
O
T
Y
O
N
E
E
T
E
W
T
N
W
Y
T
O
W
N
T
Y
T
E
T
N
N
E
By Michael Schroeder, Biotec, 2004
O
N
N
E
E
24
Dotplot of Repeats/Palindrome
M
M
A
A
M
A
M
A
A
A
D
A
M
A
M
M
A
A
D
A
M
A
M
I
A
M
M
M
M
I
M
A
M
A
D
A
A
D
A
M
M
D
A
M
I
M
D
M
D
A
M
By Michael Schroeder, Biotec, 2004
A
D
A
A
M
M
A
M
25
Dotplot for shared domain
D
D
O
O
O
T
H
Y
H
O
D
G
K
I
N
D
O
R
O
R
O
O
O
O
R
O
T
T
H
H
Y
H
Y
M
I
I
L
L
E
R
R
By Michael Schroeder, Biotec, 2004
26
Dot plot
Result
dorothycrowfoothodgkin
d*
*
o * *
* ** *
r *
*
o * *
* ** *
t
*
*
h
*
*
y
*
h
*
*
o * *
* ** *
d*
*
g
*
k
*
i
*
n
*
By Michael Schroeder, Biotec, 2004
27
Dot plot
Result
dorothycrowfoothodgkin
d*
*
o * *
* ** *
r *
*
o * *
* ** *
t
*
*
h
*
*
y
*
h
*
*
o * *
* ** *
d*
*
g
*
k
*
i
*
n
*
By Michael Schroeder, Biotec, 2004
28
Dotplots
Window size 15
Dot if
6 matches in window
By Michael Schroeder, Biotec, 2004
29
Window size 15
Dot if
6 matches in window
By Michael Schroeder, Biotec, 2004
30
>gi|1942644|pdb|1MEG|
64
Length = 216
Crystal Structure Of A Caricain D158e Mutant In Complex With E-
Score = 271 bits (693), Expect = 1e-73
Identities = 142/216 (65%), Positives = 168/216 (77%), Gaps = 4/216 (1%)
Query: 1
Sbjct: 1
Query: 61
Sbjct: 61
IPEYVDWRQKGAVTPVKNQGSCGSCWAFSAVVTIEGIIKIRTGNLNQYSEQELLDCDRRS 60
+PE VDWR+KGAVTPV++QGSCGSCWAFSAV T+EGI KIRTG L + SEQEL+DC+RRS
LPENVDWRKKGAVTPVRHQGSCGSCWAFSAVATVEGINKIRTGKLVELSEQELVDCERRS 60
YGCNGGYPWSALQLVAQYGIHYRNTYPYEGVQRYCRSREKGPYAAKTDGVRQVQPYNQGA 120
+GC GGYP AL+ VA+ GIH R+ YPY+ Q CR+++ G
KT GV +VQP N+G
HGCKGGYPPYALEYVAKNGIHLRSKYPYKAKQGTCRAKQVGGPIVKTSGVGRVQPNNEGN 120
Query: 121 LLYSIANQPVSVVLQAAGKDFQLYRGGIFVGPCGNKVDHAVAAV----GYGPNYILIKNS 176
LL +IA QPVSVV+++ G+ FQLY+GGIF GPCG KV+HAV AV
G YILIKNS
Sbjct: 121 LLNAIAKQPVSVVVESKGRPFQLYKGGIFEGPCGTKVEHAVTAVGYGKSGGKGYILIKNS 180
Query: 177 WGTGWGENGYIRIKRGTGNSYGVCGLYTSSFYPVKN 212
WGT WGE GYIRIKR GNS GVCGLY SS+YP KN
Sbjct: 181 WGTAWGEKGYIRIKRAPGNSPGVCGLYKSSYYPTKN 216
1
61
121
181
lpenvdwrkk
hgckggyppy
llnaiakqpv
wgtawgekgy
gavtpvrhqg
aleyvakngi
svvveskgrp
irikrapgns
By Michael Schroeder, Biotec, 2004
scgscwafsa
hlrskypyka
fqlykggife
pgvcglykss
vatveginki rtgklvelse qelvdcerrs
kqgtcrakqv ggpivktsgv grvqpnnegn
gpcgtkveha vtavgygksg gkgyilikns
yyptkn
31
Window size 15
Dot if
6 matches in window
By Michael Schroeder, Biotec, 2004
32
>gi|2624670|pdb|1AIM|
Fluoromethylketone
Length = 215
Cruzain Inhibited By Benzoyl-Tyrosine-Alanine-
Score = 121 bits (303), Expect = 3e-28
Identities = 78/202 (38%), Positives = 107/202 (52%), Gaps = 13/202 (6%)
Query: 2
Sbjct: 2
Query: 62
Sbjct: 62
PEYVDWRQKGAVTPVKNQGSCGSCWAFSAVVTIEGIIKIRTGNLNQYSEQELLDCDRRSY 61
P VDWR +GAVT VK+QG CGSCWAFSA+ +E
+
L
SEQ L+ CD+
PAAVDWRARGAVTAVKDQGQCGSCWAFSAIGNVECQWFLAGHPLTNLSEQMLVSCDKTDS 61
GCNGGYPWSALQLVAQY---GIHYRNTYPY---EGVQRYCRSREKGPYAAKTDGVRQVQP 115
GC+GG
+A + + Q
++ ++YPY
EG+
C +
A T V
Q
GCSGGLMNNAFEWIVQENNGAVYTEDSYPYASGEGISPPCTTSGHTVGATITGHVELPQD 121
Query: 116 YNQGALLYSIANQPVSVVLQAAGKDFQLYRGGIFVGPCGNKVDHAVAAVGYGPN----YI 171
Q A
++ N PV+V + A+
+ Y GG+
+DH V VGY +
Y
Sbjct: 122 EAQIAAWLAV-NGPVAVAVDAS--SWMTYTGGVMTSCVSEALDHGVLLVGYNDSAAVPYW 178
Query: 172 LIKNSWGTGWGENGYIRIKRGT 193
+IKNSW T WGE GYIRI +G+
Sbjct: 179 IIKNSWTTQWGEEGYIRIAKGS 200
By Michael Schroeder, Biotec, 2004
33
Window size 15
Dot if
6 matches in window
By Michael Schroeder, Biotec, 2004
34
gi|7546546|pdb|1EF7|B
Length = 242
Chain B, Crystal Structure Of Human Cathepsin X
Score = 52.0 bits (123), Expect = 2e-07
Identities = 60/231 (25%), Positives = 94/231 (40%), Gaps = 34/231 (14%)
Query: 1
Sbjct: 1
Query: 52
Sbjct: 61
IPEYVDWRQKGAV---TPVKNQ---GSCGSCWAFSAVVTIEGIIKIRTGNL---NQYSEQ 51
+P+ DWR
V
+ +NQ
CGSCWA ++
+
I I+
S Q
LPKSWDWRNVDGVNYASITRNQHIPQYCGSCWAHASTSAMADRINIKRKGAWPSTLLSVQ 60
ELLDCDRRSYGCNGGYPWSALQLVAQYGIHYRNTYPYEGVQRYCR--------SREKGPY 103
++DC
C GG
S
Q+GI
Y+
+ C
+ K +
NVIDCGNAG-SCEGGNDLSVWDYAHQHGIPDETCNNYQAKDQECDKFNQCGTCNEFKECH 119
Query: 104 AAKTDGVRQVQPYN-----QGALLYSIANQPVSVVLQAAGKDFQLYRGGIFVGPCGNK-V 157
A +
+ +V Y
+ +
AN P+S + A +
Y GGI+
+
Sbjct: 120 AIRNYTLWRVGDYGSLSGREKMMAEIYANGPISCGIMATER-LANYTGGIYAEYQDTTYI 178
Query: 158 DHAVAAVGY----GPNYILIKNSWGTGWGENGYIRI-----KRGTGNSYGV 199
+H V+ G+
G Y +++NSWG WGE G++RI
K G G Y +
Sbjct: 179 NHVVSVAGWGISDGTEYWIVRNSWGEPWGERGWLRIVTSTYKDGKGARYNL 229
By Michael Schroeder, Biotec, 2004
35
Window size 5
Dot if
2 matches in window
By Michael Schroeder, Biotec, 2004
36
Window size 1
Dot if
1 match in window
By Michael Schroeder, Biotec, 2004
37
Dynamic programming
By Michael Schroeder, Biotec, 2004
38
From Dotplots to Alignments
Obvious best alignment:
D O R O T H
D
Y C R O W F O O T H O D G K I
D
O
O
O
R
O
N
D
R
O
O
O O
O
O
O O
O
R
O
T
T
H
T
H
Y
H
Y
H
H
O
D
DOROTHYCROWFOOTHODGKIN
DOROTHY--------HODGKIN
O
O
D
G
K
I
H
O
O O
O
D
G
K
I
N
By Michael Schroeder, Biotec, 2004
N
39
From Dotplots to Alignments
Find “best” path from top left corner to bottom right
Moving “east” corresponds to “-” in the second
sequence
Moving “south” corresponds to “-” in the first
sequence
Moving “southeast” corresponds to a match if the
characters are the same or a mismatch otherwise
Can we automate this?
By Michael Schroeder, Biotec, 2004
40
From Dotplots to Alignments
Algorithm (Dynamic Programming):
Insert a row 0 and column 0 initialised with 0
Starting from the top left, move down row by row from row 1 and
right column by column from column 1 visiting each cell
Consider
The value of the cell north
The value of the cell west
The value of the cell northwest if the row/column character
mismatch
1 + the value of the cell northwest if the row/column
character match
Put down the maximum of these values as the value for the
current cell
Trace back the path with the highest values from the bottom right
to the top left and output the alignment
By Michael Schroeder, Biotec, 2004
41
From Dotplots to Alignments
0
1
T
2
G
3
C
4
A
5
T
6
A
0
1A
2T
3C
4T
5G
6A
7T
By Michael Schroeder, Biotec, 2004
42
From Dotplots to Alignments
0
0
0
1A
2T
3C
4T
5G
6A
7T
1
T
0
2
G
0
3
C
0
4
A
0
5
T
0
6
A
0
0
0
0
0
0
0
0
Insert a row 0 and column 0 initialised with 0
By Michael Schroeder, Biotec, 2004
43
From Dotplots to Alignments
0
0
1A
2T
3C
4T
5G
6A
7T
0
1
T
0
2
G
0
3
C
0
4
A
0
5
T
0
6
A
0
0
0
0
0
0
0
0
0
0
0
1
1
1
•Consider
•Value north
•Value west
•Value northwest if the row/column character mismatch
•1 + value northwest if the row/column character match
•Put down the maximum of these values for current celll
By Michael Schroeder, Biotec, 2004
44
From Dotplots to Alignments
0
0
1A
2T
3C
4T
5G
6A
7T
0
1
T
0
2
G
0
3
C
0
4
A
0
5
T
0
6
A
0
0
0
0
0
0
0
0
0
1
1
1
1
1
1
0
1
1
1
2
2
2
0
1
2
2
2
2
2
1
1
2
2
2
3
3
1
2
2
3
3
3
4
1
2
2
3
3
4
4
By Michael Schroeder, Biotec, 2004
45
Reading the Alignment
0
0
1A
2T
3C
4T
5G
6A
7T
0
1
T
0
2
G
0
3
C
0
4
A
0
5
T
0
6
A
0
0
0
0
0
0
0
0
0
1
1
1
1
1
1
0
1
1
1
2
2
2
0
1
2
2
2
2
2
1
1
2
2
2
3
3
1
2
2
3
3
3
4
1
2
2
3
3
4
4
-tgcat-aat-c-tgat
By Michael Schroeder, Biotec, 2004
46
Reading the Alignment:
there are more than one possibility
0
0
1A
2T
3C
4T
5G
6A
7T
0
1
T
0
2
G
0
3
C
0
4
A
0
5
T
0
6
A
0
0
0
0
0
0
0
0
0
1
1
1
1
1
1
0
1
1
1
2
2
2
0
1
2
2
2
2
2
1
1
2
2
2
3
3
1
2
2
3
3
3
4
1
2
2
3
3
4
4
---tgcata
atctg-atBy Michael Schroeder, Biotec, 2004
47
Formally:
Longest Common Subsequence LCS
What is the length s(V,W) of the longest common
subsequence of two sequences
V=v1..vn and W=w1..wm ?
Find sequences of indices
1 ≤ i1 < … < ik ≤ n and 1 ≤ j1 < … < jk ≤ m
such that vit = wjt for 1 ≤ t ≤ k
How? Dynamic programming:
si,0 = s0,j = 0 for all 1 ≤ i ≤ n and 1 ≤ j ≤ m and
si-1,j
si,j = max
si,j-1
si-1,j-1 + 1, if vi = wj
Then s(V,W) = sn,m is the length of the LCS
{
By Michael Schroeder, Biotec, 2004
48
Example LCS
0
1
T
2
G
3
C
4
A
5
T
6
A
0
1A
2T
3C
4T
5G
6A
7T
By Michael Schroeder, Biotec, 2004
49
Example LCS:
0
0
1A
2T
3C
4T
5G
6A
7T
0
0
0
0
0
0
0
0
1
T
0
2
G
0
3
C
0
4
A
0
5
T
0
6
A
0
Initialisation:
si,0 = s0,j = 0 for all 1 ≤ i ≤ n and 1 ≤ j ≤ m and
By Michael Schroeder, Biotec, 2004
50
Example LCS:
0
0
1A
2T
3C
4T
5G
6A
7T
0
1
T
0
2
G
0
3
C
0
4
A
0
5
T
0
6
A
0
0
0
0
0
0
0
0
0
0
0
1
1
1
Computing each cell:
si,j = max
By Michael Schroeder, Biotec, 2004
{
si-1,j
si,j-1
si-1,j-1 + 1, if vi = wj
51
Example LCS:
0
0
1A
2T
3C
4T
5G
6A
7T
0
1
T
0
2
G
0
3
C
0
4
A
0
5
T
0
6
A
0
0
0
0
0
0
0
0
0
1
1
1
1
1
1
0
1
1
1
2
2
2
0
1
2
2
2
2
2
1
1
2
2
2
3
3
1
2
2
3
3
3
4
1
2
2
3
3
4
4
Computing each cell:
si,j = max
By Michael Schroeder, Biotec, 2004
{
si-1,j
si,j-1
si-1,j-1 + 1, if vi = wj
52
LCS Algorithm
LCS(V,W)
For i = 1 to n
si,0 = 0
For j = 1 to m
s0,j = 0
For i = 1 to n
For j = 1 to m
Complexity:
LCS has quadratic complexity:
O(n m)
If vi = wj and si-1,j-1 +1 ≥ si-1,j and si-1,j-1 +1 ≥ si,j-1 Then
si,j = si-1,j-1 +1
bi,j = North West
Else if si-1,j ≥ si,j-1 Then
si,j = si-1,j
bi,j = North
Else
si,j = si,j-1
bi,j = West
Return s and b
By Michael Schroeder, Biotec, 2004
53
Printing the alignment of LCS
PRINT-LCS(b,V,i,j)
If i=0 or j=0 Then Return
If bi,j = North West Then
PRINT-LCS(V,b,i-1,j-1)
Print vi
Else if bi,j = North Then
PRINT-LCS(V,b,i-1,j)
Else
PRINT-LCS(V,b,i,j-1)
By Michael Schroeder, Biotec, 2004
54
Rewards/Penalities
We can use different schemes:
-1 for insert/delete/mismatch
+1 for match
…Consider
-1 + the value of the cell north
-1 + the value of the cell west
-1 + the value of the cell northwest if the row/column
character mismatch
+1 + the value of the cell northwest if the row/column
character match
Put down the maximum of these values as the value for
the current cell
By Michael Schroeder, Biotec, 2004
55
Reading the Alignment
0
0
1A
2T
3C
4T
5G
6A
7T
0
1
T
0
2
G
0
3
C
0
4
A
0
5
T
0
6
A
0
0
0
0
0
0
0
0
-1
1
0
1
0
-1
1
-1
0
-1
0
2
1
0
-1
-1
1
0
1
1
0
1
0
0
0
0
2
1
0
2
1
1
0
1
3
1
1
1
0
0
1
2
---tgcata
atctg-atBy Michael Schroeder, Biotec, 2004
56
Rewards/Penalities
Let’s refine the schemes:
Transition mutations are more common:
purine<->purine, a<->g
pyrimidine<->pyrimidine, t<->c
Transversions (purine<->pyrimidine) are less
common
Use a subsitutation matrix to rate mismatches:
A
A
G C
2 -2 0 -2
T -2 2 -2 0
G
-2 for insert/delete
Mismatch/match according to substitution matrix
T
0 -2 2 -2
C -2 0 -2 2
…Consider
-2 + the value of the cell north
-2 + the value of the cell west
Corresponding value of the substion matrix
+ the value of the cell northwest
Put down the maximum of these values as the
value for the current cell
By Michael Schroeder, Biotec, 2004
57
Reading the Alignment
0
0
1A
2T
3C
4T
5G
6A
7T
0
1
T
0
2
G
0
3
C
0
4
A
0
5
T
0
6
A
0
0
0
0
0
0
0
0
-2
2
0
2
0
-2
2
0
0
0
0
4
2
0
-2
0
2
0
2
2
2
2
0
0
0
0
4
2
0
4
2
2
0
2
6
2
2
2
0
2
2
4
---tgcata
atctg-atBy Michael Schroeder, Biotec, 2004
58
Substitution matrixes
By Michael Schroeder, Biotec, 2004
59
How to derive a substitution matrix for
amino acids?
Amino acids can be classified by physiochemical
properties
Hydrophobic
A
G
P
I
L
Polar
S
T
N
Q
V
Acidic
D
E
Y
C
W
M
F
By Michael Schroeder, Biotec, 2004
K
H
R Basic
Aromatic
60
Cys
Ser
Thr
Pro
Ala
Gly
Asn
Asp
Glu
Gln
His
Arg
Lys
Met
Ile
Leu
Val
Phe
Tyr
Trp
12
0
-2
-3
-2
-3
-4
-5
-5
-5
-3
-4
-5
-5
-2
-6
-2
-4
0
-8
C
PAM 250 matrix
2
1
1
1
1
1
0
0
-1
-1
0
0
-2
-1
-3
-1
-3
-3
-2
S
3
0
1
0
0
0
0
-1
-1
-1
0
-1
0
-2
0
-3
-3
-5
T
6
1
-1
-1
-1
-1
0
0
0
-1
-2
-2
-3
-1
-5
-5
-6
P
2
1
0
0
0
0
-1
-2
-1
-1
-1
-2
0
-4
-3
-6
A
5
0
1
0
-1
-2
-3
-2
-3
-3
-4
-1
-5
-5
-7
G
2
2
1
1
2
0
1
-2
-2
-3
-2
-4
-2
-4
N
By Michael Schroeder, Biotec, 2004
4
3
2
1
-1
0
-3
-2
-4
-2
-6
-4
-7
D
>0, likely mutation
0, random mutation
<0, unlikely
4
2
1
-1
0
-2
-2
-3
-2
-5
-4
-7
E
4
3
1
1
-1
-2
-2
-2
-5
-4
-5
Q
6
2
0
-2
-2
-2
-2
-2
0
-3
H
6
3
0
-2
-3
-2
-4
-4
2
R
5
0 6
-2 2 5
-3 4 2
-2 2 4
-5 0 1
-4 -2 -1
-3 -4 -5
K M I
6
2 4
2 -1 9
-1 -2 7 10
-2 -6 0 0 17
L V F Y W
61
Cys
12
0
-2
-3
-2
-3
-4
-5
-5
-5
-3
-4
-5
-5
-2
-6
-2
-4
0
-8
Ser
0
2
1
1
1
1
1
0
0
-1
-1
0
0
-2
-1
-3
-1
-3
-3
-2
Thr
-2
1
3
0
1
0
0
0
0
-1
-1
-1
0
-1
0
-2
0
-3
-3
-5
Pro
-3
1
0
6
1
-1
-1
-1
-1
0
0
0
-1
-2
-2
-3
-1
-5
-5
-6
Ala
-2
1
1
1
2
1
0
0
0
0
-1
-2
-1
-1
-1
-2
0
-4
-3
-6
Gly
-3
1
0
-1
1
5
0
1
0
-1
-2
-3
-2
-3
-3
-4
-1
-5
-5
-7
Asn
-4
1
0
-1
0
0
2
2
1
1
2
0
1
-2
-2
-3
-2
-4
-2
-4
Asp
-5
0
0
-1
0
1
2
4
3
2
1
-1
0
-3
-2
-4
-2
-6
-4
-7
Glu
-5
0
0
-1
0
0
1
3
4
2
1
-1
0
-2
-2
-3
-2
-5
-4
-7
Gln
-5
-1
-1
0
0
-1
1
2
2
4
3
1
1
-1
-2
-2
-2
-5
-4
-5
His
-3
-1
-1
0
-1
-2
2
1
1
3
6
2
0
-2
-2
-2
-2
-2
0
-3
Arg
-4
0
-1
0
-2
-3
0
-1
-1
1
2
6
3
0
-2
-3
-2
-4
-4
2
Lys
-5
0
0
-1
-1
-2
1
0
0
1
0
3
5
0
-2
-3
-2
-5
-4
-3
Met
-5
-2
-1
-2
-1
-3
-2
-3
-2
-1
-2
0
0
6
2
4
2
0
-2
-4
Ile
-2
-1
0
-2
-1
-3
-2
-2
-2
-2
-2
-2
-2
2
5
2
4
1
-1
-5
Leu
-6
-3
-2
-3
-2
-4
-3
-4
-3
-2
-2
-3
-3
4
2
6
2
2
-1
-2
Val
-2
-1
0
-1
0
-1
-2
-2
-2
-2
-2
-2
-2
2
4
2
4
-1
-2
-6
Phe
-4
-3
-3
-5
-4
-5
-4
-6
-5
-5
-2
-4
-5
0
1
2
-1
9
7
0
Tyr
0
-3
-3
-5
-3
-5
-2
-4
-4
-4
0
-4
-4
-2
-1
-1
-2
7
10
0
Trp
-8
-2
-5
-6
-6
-7
-4
-7
-7
-5
-3
2
-3
-4
-5
-2
-6
0
0
17
Average
StDev
C
S
T
P
A
G
N
D
E
Q
H
R
K
M
I
L
V
F
Y
W
-2.8
-0.5
-0.7
-1.2
-0.9
-1.6
-0.7
-1.1
-1.1
-0.8
-0.3
-0.7
-0.9
-0.8
-0.8
-1.4
-0.8
-1.9
-1.5
-3.1
4
1.5
1.7
2.6
1.9
2.7
2.1
3
2.8
2.6
2.3
2.6
2.5
2.6
2.4
3
2.3
4.1
3.8
5.4
By Michael Schroeder, Biotec, 2004
62
Cys
12
0
-2
-3
-2
-3
-4
-5
-5
-5
-3
-4
-5
-5
-2
-6
-2
-4
0
-8
Ser
0
2
1
1
1
1
1
0
0
-1
-1
0
0
-2
-1
-3
-1
-3
-3
-2
Thr
-2
1
3
0
1
0
0
0
0
-1
-1
-1
0
-1
0
-2
0
-3
-3
-5
Pro
-3
1
0
6
1
-1
-1
-1
-1
0
0
0
-1
-2
-2
-3
-1
-5
-5
-6
Ala
-2
1
1
1
2
1
0
0
0
0
-1
-2
-1
-1
-1
-2
0
-4
-3
-6
Gly
-3
1
0
-1
1
5
0
1
0
-1
-2
-3
-2
-3
-3
-4
-1
-5
-5
-7
Asn
-4
1
0
-1
0
0
2
2
1
1
2
0
1
-2
-2
-3
-2
-4
-2
-4
Asp
-5
0
0
-1
0
1
2
4
3
2
1
-1
0
-3
-2
-4
-2
-6
-4
-7
Glu
-5
0
0
-1
0
0
1
3
4
2
1
-1
0
-2
-2
-3
-2
-5
-4
-7
Gln
-5
-1
-1
0
0
-1
1
2
2
4
3
1
1
-1
-2
-2
-2
-5
-4
-5
His
-3
-1
-1
0
-1
-2
2
1
1
3
6
2
0
-2
-2
-2
-2
-2
0
-3
Arg
-4
0
-1
0
-2
-3
0
-1
-1
1
2
6
3
0
-2
-3
-2
-4
-4
2
Lys
-5
0
0
-1
-1
-2
1
0
0
1
0
3
5
0
-2
-3
-2
-5
-4
-3
Met
-5
-2
-1
-2
-1
-3
-2
-3
-2
-1
-2
0
0
6
2
4
2
0
-2
-4
Ile
-2
-1
0
-2
-1
-3
-2
-2
-2
-2
-2
-2
-2
2
5
2
4
1
-1
-5
Leu
-6
-3
-2
-3
-2
-4
-3
-4
-3
-2
-2
-3
-3
4
2
6
2
2
-1
-2
Val
-2
-1
0
-1
0
-1
-2
-2
-2
-2
-2
-2
-2
2
4
2
4
-1
-2
-6
Phe
-4
-3
-3
-5
-4
-5
-4
-6
-5
-5
-2
-4
-5
0
1
2
-1
9
7
0
Tyr
0
-3
-3
-5
-3
-5
-2
-4
-4
-4
0
-4
-4
-2
-1
-1
-2
7
10
0
Trp
-8
-2
-5
-6
-6
-7
-4
-7
-7
-5
-3
2
-3
-4
-5
-2
-6
0
0
17
Average
StDev
C
S
T
P
A
G
N
D
E
Q
H
R
K
M
I
L
V
F
Y
W
-2.8
-0.5
-0.7
-1.2
-0.9
-1.6
-0.7
-1.1
-1.1
-0.8
-0.3
-0.7
-0.9
-0.8
-0.8
-1.4
-0.8
-1.9
-1.5
-3.1
4
1.5
1.7
2.6
1.9
2.7
2.1
3
2.8
2.6
2.3
2.6
2.5
2.6
2.4
3
2.3
4.1
3.8
5.4
By Michael Schroeder, Biotec, 2004
63
PAM 250: Interpretation
Immutable:
Cysteine (Avg=-2.8): known to have several unique,
indispensable functions
attachment site of heme group in cytochrome and of iron
sulphur FeS in ferredoxins
Cross links in proteins such as chymotrypsin or ribonuclease
Seldom without unique function
Glycine (Avg=-1.6): small size maybe advantageous
Mutable:
Serine often functions in active site, but can be easily replaced
Self-alignment:
Tryptophan with itself scores very high, as W occurs rarely
By Michael Schroeder, Biotec, 2004
64
Point Accepted Mutations PAM
Substitution matrix using explicit evolutionary model of how
amino acids change over time
Use parsimony method to determine frequency of mutations
Entry in PAM matrix: Likelihood ratio for residues a and b: Probability
a-b is a mutation / probability a-b is chance
PAM x: Two sequences V, W have evolutionary distance of x PAM if
a series of accepted point mutations (and no insertions/deletions)
converts V into W averaging to x point mutation per 100 residues
Mutations here = mutations in the DNA
Because of silent mutations and back mutations n can be >100
PAM 250 most commonly used
By Michael Schroeder, Biotec, 2004
65
PAM and Sequence Similarity
PAM
0
30
80
110
200
250
% identiy
100
75
60
50
25
20
By Michael Schroeder, Biotec, 2004
66
PAM
Dayhoff, Eck, Park: A model of evolutionary change in proteins,
1978
Accepted point mutation = substitution of an amino acid
accepted by natureal selection
Assumption: X replacing Y as likely as Y replacing X
Used cytochrome c, hemoglobin, myoglobin, virus coat
proteins, chymotrypsinogen, glyceraldehyde 3-phosphate
dehrydogenase, clupeine, insulin, ferredoxin
Sequences which are too distantly related have been
omitted as they are more likely to contain multiple mutations
per site
By Michael Schroeder, Biotec, 2004
67
PAM: Step 1
Step 1: Construct a multiple alignment
Example
ACGCTAFKI
GCGCTAFKI
ACGCTAFKL
GCGCTGFKI
GCGCTLFKI
ASGCTAFKL
ACACTAFKL
By Michael Schroeder, Biotec, 2004
68
PAM: Step 2
Create a phylogenetic tree (parsimony method)
ACGCTAFKI
A->G
I->L
GCGCTAFKI
A->G
GCGCTGFKI
ACGCTAFKL
A->L
GCGCTLFKI
By Michael Schroeder, Biotec, 2004
C->S
ASGCTAFKL
G->A
ACACTAFKL
69
PAM: Step 3
Note, the following variables
Residue frequency ri is the number of amino acid i
occurring in the sequences, e.g. rA = 10 and rG=10
Number of residues r is the number of overall amino
acids in all sequences, e.g. r=63
Substitutability si is the number of substitutions in the
tree involving amino acid i , e.g. sA=4
Substituion frequency si,j is the number of
substitutions involving amino acid i and j (i.e. the
number of ij and ji ), e.g. sA,G = 3
Number of substitutions s is the number of overall
substitutions, s=6
By Michael Schroeder, Biotec, 2004
70
PAM: Step 4
Relative mutability is the number of times the amino
acid is substituted by any other amino acid in the tree
divided by the total number of substitutions that could
have affected the residue
Note, it is assumed that substitutions in both
directions are equally likely
How many residues, how
many substitutions overall?
mi = 100 x ( si x ri ) / ( 2 s x r )
How many i? How many
substitutions with i?
Example mA = 100 x ( 4 x 10 ) / ( 2 x 6 x 63 ) = 5.3
By Michael Schroeder, Biotec, 2004
71
PAM: Step 5
Compute mutation probability
Mi,j = mj x si,j / sj
Example: MG,A = 5.3 x 3 / 4 = 3.975
By Michael Schroeder, Biotec, 2004
72
PAM: Step 6
Finally the entry in the PAM Matrix:
Ri,j = log ( Mi,j / ( ri / r ) )
Example: RG,A = log ( 3.975 / (10/63) ) = 1.4
By Michael Schroeder, Biotec, 2004
73
PAM: Step 6
For the entries on the diagonal
mj =relative mutability 1-mj relative immutability
Rj,j = log ( (1-mj ) / ( rj / r ) )
By Michael Schroeder, Biotec, 2004
74
BLOSUM
Different approach to PAM
BLOcks SUbstitution Matrix (based on BLOCKS
database)
Generation of BLOSUM x
Group highly similar sequences and replace them by a
representative sequences.
Only consider sequences with no more than x % similarity
Align sequences (no gaps)
For any pair of amino acids a,b and for all columns c of the
alignment, let q(a,b) be the number of co-occurrences of a,b in all
columns c.
Let p(a) be the overall probability of a occurring
BLOSUM entry for a,b is log2 ( q(a,b) / ( p(a)*p(b) ) )
BLOSUM 50 and BLOSUM 62 widely used
By Michael Schroeder, Biotec, 2004
75
LCS Algorithm (Longest Common
Subsequence) Revisited
Algorithm (Dynamic Programming) with Substitution Matrix:
Insert a row 0 and column 0 initialised with 0
Starting from the top left, move down row by row from row 1 and
right column by column from column 1 visiting each cell
Consider
The value of the cell north
The value of the cell west
The value of the cell northwest if the row/column character
mismatch
s + the value of the cell northwest, where s is the value
in the subsitution matrix for the residues in row/column
Put down the minimum of these values as the value for the
current cell
Trace back the path with the highest values from the bottom right
to the top left and output the alignment
By Michael Schroeder, Biotec, 2004
76
LCS Revisited: Formally
What is the length s(V,W) of the longest common subsequence
of two sequences
V=v1..vn and W=w1..wm ?
Find sequences of indices
1 ≤ i1 < … < ik ≤ n and 1 ≤ j1 < … < jk ≤ m
such that vit = wjt for 1 ≤ t ≤ k
How? Dynamic programming:
si,0 = s0,j = 0 for all 1 ≤ i ≤ n and 1 ≤ j ≤ m and
si-1,j
si,j = max
si,j-1
si-1,j-1 + t, where t is the value for vi and wj in
the substitution matrix
Then s(V,W) = sn,m is the length of the LCS
{
By Michael Schroeder, Biotec, 2004
77
Dynamic programming revisited:
local and global alignments and gap
By Michael Schroeder, Biotec, 2004
78
Evolution and Alignments
Alignments can be interpreted in evolutionary terms
Identical letters are aligned.
Interpretation: part of the same ancestral sequence
and not changed
Non-identical letters are aligned (substitution)
Interpretation: Mutation
Gaps
Interpretation: Insertions and deletions (indels)
By Michael Schroeder, Biotec, 2004
79
Evolution and Alignments
Specific problems aligning DNA:
“Frame shift”:
DNA triplets code amino acids
Indel of one nucleotide shifts the whole sequence of
triplets
Thus may have a global effect and change all coded
amino acids
Silent mutation:
Substitution in DNA leaves transcribed amino acid
unchanged
Non-sense mutation:
Substitution to stop-codon
By Michael Schroeder, Biotec, 2004
80
Local and Global Alignments
Global alignment (Needleham-Wunsch) algorithm finds
overall best alignment
Example: members of a protein family, e.g. globins are very
conserved and have the same length in different organisms from
fruit fly to humans
Local alignment (Smith-Waterman) algorithm finds locally best
alignment
most widely used, as
e.g. genes from different organisms retain similar
exons, but may have different introns
e.g. homeobox gene, which regulates embryonic
development occurs in many species, but very different
apart from one region called homeodomain
e.g. proteins share some domains, but not all
By Michael Schroeder, Biotec, 2004
81
Local Alignment
LCS s(V,W) computes globally best alignment
Often it is better to maximise locally, i.e. compute
maximal s(vi…vi’ , wj… wi’ ) for all substrings of V and
W
Can we adapt algorithm?
Global alignment = longest path in matrix s from (0,0)
to (n,m)
Local alignment = longest path in matrix s from any
(i,j) to any (i’,j’)
Modify definition of s adding vertex of weight 0 from
source to every other vertex, creating a free “jump” to
any starting position (i,j)
By Michael Schroeder, Biotec, 2004
82
Local Alignment
Modify the definition of s as follows:
si,0 = s0,j = 0 for all 1 ≤ i ≤ n and 1 ≤ j ≤ m and
0
si-1,j
si,j = max
si,j-1
si-1,j-1 + t, where t is the value for vi wj
in the substitution matrix
Then s(V,W) = max { si,j } is the length of the local LCS
{
This computes longest path in edit graph
Several local alignment may have biological
significance (consider e.g. two multi-domain proteins
whose domains are re-ordered
By Michael Schroeder, Biotec, 2004
83
Aligning with Gap Penalties
Gap is sequence of spaces in alignment
So far, we consider only insertion and deletion of single
nucleotides or amino acids creating alignments with many gaps
So far, score of a gap of length l is l
Because insertion/deletion of monomers is evolutionary slow
process, large numbers of gaps do not make sense
Instead whole substrings will be deleted or inserted
We can generalise score of a gap to a score function A + B l,
where A is the penalty to open the gap and B is the penalty
to extend the gap
By Michael Schroeder, Biotec, 2004
84
Aligning with Gap Penalties
High gap penalties result in shorter, lower-scoring
alignments with fewer gaps and
Lower gap penalties give higher-scoring, longer
alignments with more gaps
Gap opening penalty A mainly influences number
of gaps
Gap extension penalty B mainly influences length
of gaps
E.g. if interested in close relationships, then choose
A, B above default values, for distant relationships
decrease default values
By Michael Schroeder, Biotec, 2004
85
Aligning with Gap Penalties
Adapt the definition of s as follows:
s-deli,j = max
s-insi,j = max
{
{
s-deli-1,j - B
si-1,j – (A+B)
s-insi,j-1 - B
si,j-1 – (A+B)
0
s-deli,j
si,j = max
s-insi,j
si-1,j-1 + t, where t is the value for vi, wj
in the substitution matrix
Then s(V,W) = max { si,j } is the length of the local LCS
with gap penalties A and B
{
By Michael Schroeder, Biotec, 2004
86
FASTA and BLAST
By Michael Schroeder, Biotec, 2004
87
Motivation
As in dotplots, the underlying data structure for dynamic
programming is a table
Given two sequences of length n dynamic programming
takes time proportional to n2
Given a database with m sequences, comparing a query
sequence to the whole database takes time proportional to m
n2
What does this mean?
Imagine you need to fill in the tables by hand and it takes 10
second to fill in one cell
Assume there are 1.000.000 sequences each 100 amino acids
long
How long does it take?
By Michael Schroeder, Biotec, 2004
88
1.000.000 x 100 x 100 x 10 sec = 1011 sec =
27.777.778h = 1157407days = 3170 years
Even if a computer does not take 10 sec, but just
0.1ms to fill in one cell, it would still be 12 days.
We cannot do something about the database size,
but can we do something about the table size?
By Michael Schroeder, Biotec, 2004
89
An idea: Prune the search space
By Michael Schroeder, Biotec, 2004
90
Another idea
Did we formulate the
problem correctly?
Do we need the alignments
for all sequences in the
database?
No, only for “reasonable”
hits introduce a
threshold
A “reasonable” alignment will
contain short stretches of
perfect matches
Find these first, then
extend them to connect
them as best possible
By Michael Schroeder, Biotec, 2004
91
FASTA and BLAST
FASTA and BLAST faster than dynamic programming
(5 times and 50 times respectively)
Underlying idea for a heuristic:
High-scoring alignments will contain short stretches
of identical letters, called words
FASTA and BLAST first search for matches of words of
a given length and score threshold:
BLAST for words of length 3 for proteins and 11 for
DNA
FASTA for words of length 2 for proteins and 6 for
DNA
Next, matches are extended to local (BLAST) and
global (FASTA) alignments
By Michael Schroeder, Biotec, 2004
92
FASTA and BLAST
More formally:
If the strings V=v1..vm and W=w1..wm match with at
most k mismatches, then they share an p-tuple for
p = m/(k+1),
i.e. vi..vi+l-1 =wj..wj+l-1 for some 1 ≤ i,j ≤ m-p+1
FILTRATION ALGORITHM, which detects all matching
words of length m with up to k mismatches
Potential match detection: Find all matches of p-tuples
of V,W (can be done in linear time by inserting them into
a hash table)
Potential match verification: Verify each potential
match by extending it to the left and right until either the
first k+1 mismatches are found or the beginning or end
of the sequences are found
By Michael Schroeder, Biotec, 2004
93
Example for BLAST
Search SWISSPROT for Immunoglobulin:
SWISS_PROT:C79A_HUMAN P11912
By Michael Schroeder, Biotec, 2004
94
Example for BLAST
Search BLAST (www.ncbi.nlm.nih.gov/BLAST/) for P11912
Database: All non-redundant SwissProt sequences
1,292,592 sequences; 412,925,052 total letters
By Michael Schroeder, Biotec, 2004
95
Example for BLAST
Distribution of Hits:
By Michael Schroeder, Biotec, 2004
96
Example for BLAST: Top Hits
Score E Sequences producing significant alignments:
Score
gi|547896|sp|P11912|C79A_HUMAN B-cell antigen receptor comp...
473
gi|728993|sp|P40293|C79A_BOVIN B-cell antigen receptor comp...
312
gi|126779|sp|P11911|C79A_MOUSE B-cell antigen receptor comp...
278
gi|728994|sp|P40259|C79B_HUMAN B-cell antigen receptor comp...
55
gi|125781|sp|P01618|KV1_CANFA IG KAPPA CHAIN V REGION GOM
38
gi|125361|sp|P17948|VGR1_HUMAN Vascular endothelial growth ...
37
gi|549319|sp|P35969|VGR1_MOUSE Vascular endothelial growth ...
36
gi|114764|sp|P15530|C79B_MOUSE B-cell antigen receptor comp...
36
gi|1718161|sp|P53767|VGR1_RAT Vascular endothelial growth f...
35
gi|125735|sp|P01681|KV01_RAT Ig kappa chain V region S211
35
gi|1730075|sp|P01625|KV4A_HUMAN IG KAPPA CHAIN V-IV REGION LEN 34
gi|1718188|sp|P52583|VGR2_COTJA Vascular endothelial growth...
33
gi|125833|sp|P06313|KV4B_HUMAN IG KAPPA CHAIN V-IV REGION J... 33
gi|125806|sp|P01658|KV3F_MOUSE IG KAPPA CHAIN V-III REGION ...
33
gi|125808|sp|P01659|KV3G_MOUSE IG KAPPA CHAIN V-III REGION ...
33
gi|1172451|sp|Q05793|PGBM_MOUSE Basement membrane-specific ...
33
gi|125850|sp|P01648|KV5O_MOUSE Ig kappa chain V-V region HP...
33
gi|125830|sp|P06312|KV40_HUMAN Ig kappa chain V-IV region p...
33
gi|2501738|sp|Q06639|YD03_YEAST Putative 101.7 kDa transcri...
33
By Michael Schroeder, Biotec, 2004
E-Value
e-133
3e-85
5e-75
1e-07
0.019
0.042
0.052
0.064
0.080
0.095
0.26
0.28
0.30
0.30
0.30
0.33
0.36
0.38
0.41
97
Example for BLAST: Alignment
>gi|126779|sp|P11911|C79A_MOUSE
B-cell antigen receptor complex
associated protein alpha-chain
precursor (IG-alpha) (MB-1 membrane glycoprotein)
(Surface-IGM-associated protein) (Membrane-bound
immunoglobulin associated protein) (CD79A)
Length = 220
Score = 278 bits (711), Expect = 5e-75
Identities = 150/226 (66%), Positives = 165/226 (73%), Gaps = 6/226 (2%)
Query: 1
Sbjct: 1
Query: 61
Sbjct: 56
MPGGPGVLQALPATIFLLFLLSAVYLGPGCQALWMHKVPASLMVSLGEDAHFQCPHNSSN 60
MPGG
+ LL LS
LGPGCQAL +
P SL V+LGE+A
C N+
MPGG----LEALRALPLLLFLSYACLGPGCQALRVEGGPPSLTVNLGEEARLTC-ENNGR 55
NANVTWWRVLHGNYTWPPEFLGPGEDPNGTLIIQNVNKSHGGIYVCRVQEGNESYQQSCG 120
N N+TWW L N TWPP LGPG+
G L
VNK+ G
C+V E N
++SCG
NPNITWWFSLQSNITWPPVPLGPGQGTTGQLFFPEVNKNTGACTGCQVIE-NNILKRSCG 114
Query: 121 TYLRVRQPPPRPFLDMGEGTKNRIITAEGIILLFCAVVPGTLLLFRKRWQNEKLGLDAGD 180
TYLRVR P PRPFLDMGEGTKNRIITAEGIILLFCAVVPGTLLLFRKRWQNEK G+D D
Sbjct: 115
TYLRVRNPVPRPFLDMGEGTKNRIITAEGIILLFCAVVPGTLLLFRKRWQNEKFGVDMPD 174
Query: 181 EYEDENLYEGLNLDDCSMYEDISRGLQGTYQDVGSLNIGDVQLEKP 226
+YEDENLYEGLNLDDCSMYEDISRGLQGTYQDVG+L+IGD QLEKP
Sbjct: 175
DYEDENLYEGLNLDDCSMYEDISRGLQGTYQDVGNLHIGDAQLEKP 220
By Michael Schroeder, Biotec, 2004
98
Example for BLAST
Lineage Report
root
. cellular organisms
. . Eukaryota
[eukaryotes]
. . . Fungi/Metazoa group [eukaryotes]
. . . . Bilateria
[animals]
. . . . . Coelomata
[animals]
. . . . . . Gnathostomata
[vertebrates]
. . . . . . . Tetrapoda
[vertebrates]
. . . . . . . . Amniota
[vertebrates]
. . . . . . . . . Eutheria
[mammals]
. . . . . . . . . . Homo sapiens (man) ---------------------. . . . . . . . . . Bos taurus (bovine) .....................
. . . . . . . . . . Mus musculus (mouse) ....................
. . . . . . . . . . Canis familiaris (dogs) .................
. . . . . . . . . . Rattus norvegicus (brown rat) ...........
. . . . . . . . . . Oryctolagus cuniculus (domestic rabbit) .
. . . . . . . . . Coturnix japonica ------------------------. . . . . . . . . Gallus gallus (chickens) ..................
. . . . . . . . Xenopus laevis (clawed frog) ---------------. . . . . . . Heterodontus francisci -----------------------. . . . . . Drosophila melanogaster ------------------------. . . . . Caenorhabditis elegans ---------------------------. . . . Saccharomyces cerevisiae (brewer's yeast) ----------. . . Marchantia polymorpha --------------------------------. . Agrobacterium tumefaciens str. C58 ---------------------. Human adenovirus type 3 ----------------------------------. Human adenovirus type 7 ...................................
By Michael Schroeder, Biotec, 2004
473 33 hits [mammals]
312 2 hits [mammals]
278 31 hits [mammals]
37 1 hit [mammals]
35 7 hits [mammals]
29 1 hit [mammals]
33 2 hits [birds]
31 4 hits [birds]
30 2 hits [amphibians]
28 1 hit [sharks and rays]
30 2 hits [flies]
29 1 hit [nematodes]
33 1 hit [ascomycetes]
29 1 hit [liverworts]
28 1 hit [a-proteobacteria]
30 1 hit [viruses]
30 1 hit [viruses]
99
How good is an alignment?
Be careful: Fitch/Smith found 17 alignments for alpha- and
beta-chains in chicken haemoglobins
Only one is the correct one (according to the structure)
Given an alignment, how good is it :
Percentage of matching residues, i.e. number of matches divided
by length of smallest sequence
Advantage: independent of sequence length
E.g.
AT–C –TGAT
4/6 = 66.67%
–TGCAT –A–
More general: also consider gaps, extensions,…
By Michael Schroeder, Biotec, 2004
100
Blast Raw Score
R = a I + b X - c O - d G, where
I is the number of identities in the alignment and a is
the reward for each identity
X is the number of mismatches in the alignment and
b is the “reward” for each mismatch
O is the number of gaps and c is the penalty for each
gap
G is the number of “-” characters in the alignment
and d is the penalty for each
The values for a,b,c,d appear at the bottom of a Blast
report. For BLASTn they are a=1, b=-3, c=5, d=2
By Michael Schroeder, Biotec, 2004
101
Example
Query: 1 atgctctggccacggcacttgcgga
||||||||||||||| |||| |||
Sbjt:107 atgctctggccacggatcttgtgga
tcccagggtgatctgtgcacctgcgata 53
|||||
|||| |||||||||||||||
tccca---tgatatgtgcacctgcgata 156
R = 1 x 46 + -3 x 4 - 5 x 1 - 2 x 3 = 23
So, given the scores: how significant is the
alignment?
By Michael Schroeder, Biotec, 2004
102
Significance of an alignment
Significance of an alignment needs to be defined with respect
to a control population
Pairwise alignment: How can we get control population?
Generate sequences randomly? Not a good model of real
sequences
Chop up both sequences and randomly reassemble them
Database search: How can we get control population?
Control = whole database
Align sequence to control population and see how good result
is in comparison
This is captured by Z scores, P-values and E-values
By Michael Schroeder, Biotec, 2004
103
Z-score
Z-score normalises the score S:
Let m be mean of population and std its standard
deviation, then Z-score = (S – m) / std
Z-score of 0 no better than average, hence might
have occurred by chance
The higher the Z-score the better
By Michael Schroeder, Biotec, 2004
104
P-value
P-value: probability of obtaining a score ≥ S
Range: 0 ≤ P ≤ 1
Let m be the number of sequences in the control
population with score ≥ S
Let p be the size of the control population
Then P-value = m / p
Rule of thumb:
P ≤ 10-100 exact match,
10-100 ≤ P ≤ 10-50 nearly identical (SNPs)
10-50 ≤ P ≤ 10-10 homology certain
10-5 ≤ P ≤ 10-1 usually distant relative
P > 10-1 probably insignificant
By Michael Schroeder, Biotec, 2004
105
E-values
E-value takes also the database into account
E-value = expected frequency of a score ≥ S
Range: 0 ≤ E ≤ m, where m is the size of the database
Relationship to P: E = m P
E values are calculated from
the bit score
the length of the query
the size of the database
By Michael Schroeder, Biotec, 2004
106
BLAST Bit score
The bit score normalizes the raw score S to make
score under different settings comparable
The bit score is obtained from the raw score as
follows
S = ( lambda x R - ln(K) ) / ( ln(2),
where lambda = 1.37 and K=0.711
Example
S = ( 1.37 x 23 - ln(0.711) ) / ln(2) = 46
By Michael Schroeder, Biotec, 2004
107
E-value
The E-value is then calculated as follows:
E = m x n x 2 -S , where
m is the effective length of the query
n is the effective length of the database
S is the bit score
(effective length takes into account that an alignment
cannot start at the end of a sequence)
Example:
m=34 (19 nucleotides fewer than the 53 submitted)
n=5,854,611,841
Result: E=0.003
By Michael Schroeder, Biotec, 2004
108
Precision and Recall
How good are BLAST and FASTA?
True positives, tp = hits which are biologically meaningful
False positives, fp = hits which are not biologically meaningful
True negatives, tn = non-hits which are not biologically meaningful
False negatives, fn = non-hits which are biologically meaningful
Minimise fp and fn
Recall:
tp/(tp+fn)
Precision: tp/(tp+fp)
(meaningful hits / all meaningful)
(meaningful hits / all hits)
But: since no objective data available difficult to judge BLAST
and FASTA’s sensitivity and specificity
By Michael Schroeder, Biotec, 2004
109
Multiple Sequence Alignments
By Michael Schroeder, Biotec, 2004
110
Multiple Sequence Alignment
Align more than two sequences
Choice of sequences
If too closely related then large redundant
If very distantly related then difficult to generate good alignment
Additionally use colour for residues with similar properties
Yellow
Green
Small polar
Hydrophobic
Magenta
Red
Blue
Polar
Negatively charged
Positively charged
By Michael Schroeder, Biotec, 2004
GLy,Ala,Ser,Thr
Cys,Val,Ile,Leu,
Pro,Phe,Tyr,Met,Trp
Asn,Gln,His
Asp,Glu
Lys, Arg
111
Thioredoxins: WCGPC[K or R] motif
By Michael Schroeder, Biotec, 2004
112
Thioredoxins: Gly/Pro = turn
By Michael Schroeder, Biotec, 2004
113
Thioredoxins: every second
hydrophobic = beta strand
By Michael Schroeder, Biotec, 2004
114
Thioredoxins:
ca. every 4th hydrophobic = alpha helix
By Michael Schroeder, Biotec, 2004
115
By Michael Schroeder, Biotec, 2004
116
Profiles, PSI-Blast, HMM
By Michael Schroeder, Biotec, 2004
117
Profiles
Derive profile from multiple sequence alignment
Useful to
Align distantly related sequences
Conserved regions, which may indicate active site
Classify subfamilies within homologues
How can profile be used to search
Insist on profile (such as WGCPC)? Too strict
Use frequence distribution of profile…
By Michael Schroeder, Biotec, 2004
118
Consider frequencies
A
25
C D
E
F
G H
1
I
K
L
M N
P
Q R
T
2
26
V
W Y
13
16
27
16
28
29
S
7
1
7
1
5 3
16
30
1
4
2
1
Score for
VDFSAS = 13+16+16+7+16+7
ADATAA = 1+16+0+1+16+0
Not good to pick up distant relationships
Better: combine with substitution matrix
Result: position specific substitution matrix
By Michael Schroeder, Biotec, 2004
119
PSI-Blast
Globin familiy (oxygen transport ) of proteins occurs
in many species
Proteins have same function and structure and
But there are pairs of members of the family sharing
less than 10% identical residues
A
50%
50%
B
C
Only 10%
PSI-BLAST idea: score via intermediaries may be
better than score from direct comparison
By Michael Schroeder, Biotec, 2004
120
PSI-BLAST
PSI-BLAST
1. BLAST
2. Collect top hits
3. Build multiple sequence alignment from significant
local matches
4. Build profile
5. Re-probe database with profile
6. Go back to 2.
By Michael Schroeder, Biotec, 2004
121
PSI-BLAST
But beware of PSI-BLAST:
False positives propagate and spread through
iterations
If protein A consists of domains D and E, and protein B
of domains E and F and protein C of domain F, then
PSI-BLAST will relate A and C although they do not
share any domain
By Michael Schroeder, Biotec, 2004
122
Hidden Markov Model
Procedure to generate sequences
State transition systems with three types of states
Deletion
Insertion
Match, which emits residues
Follow probability distribution for successor state
Train model on multiple sequence alignment
del
start
ins
match
By Michael Schroeder, Biotec, 2004
del
del
ins
ins
match
match
end
123
Summary
Evolutionary model: Indels and substitutions
Homologues vs. similarity
Dot plots
Easy visual exploration, but not scalable
Dynamic programming
Local, global, gaps
Substitution matrices (PAM, BLOSUM)
BLAST and FASTA
Scores and significance
Multiple Sequence Alignments
Profiles, PSI-BLAST, HMM
By Michael Schroeder, Biotec, 2004
124
Phylogeny
By Michael Schroeder, Biotec, 2004
125
Motivation
How did the nucleus get into the eucaryotic cells?
From where are we?
Recent Africa vs. Multi-regional Hypothese
In 1999 Encephalitis caused by the West Nile Virus
broke out in New York. How did the virus come to
New York?
By Michael Schroeder, Biotec, 2004
126
How did the nucleus get into the
eucaryotic cells?
Simple experiment:
Blast classes genes
with related functions
in yeast (Eucaryote)
against Bacteria
and
against Archaea
And count number of
significant hits
By Michael Schroeder, Biotec, 2004
127
How did the nucleus get into the
eucaryotic cells?
Mitochondria und
Energy metabolism:
Significantly more hits
in bacteria
Cell organisation:
Significantly more hits
in Archaea
Blue = Bacteria
Grey = Archaea
Fundamental Result
without any experiment!
By Michael Schroeder, Biotec, 2004
128
Phylogeny
Taxonomists aim to classify and group
organisms
E.g. Aristoteles, De Partibus Animalium
Ought we, for instance, to begin by discussing
each separate species – man, lion, ox, and the
like – taking each kind in hand independently of
the rest, or ought we rather to deal first with the
attributes which they have in common in virtue
of some common element of their nature, and
proceed from this as a basis for consideration of
them separately other
By Michael Schroeder, Biotec, 2004
129
Schools of Taxonomists
Goal: create taxonomy
Approach:
Phenotype
Phylogeny
3 schools:
Phenotype only
Evolutionary
Taxonomists:
Phenotype (+ Phylogeny)
Cladists:
Phylogeny (+Phenotype)
By Michael Schroeder, Biotec, 2004
130
Practical Application:
Westnile virus in NY
Westnile virus mainly in
Africa
Transmitted by insects and
birds
How did the virus get to NY
in 1999
Hundreds of DNA samples
taken
All 99.8% identical
single entry to NY!
Phylogenetic tree allows to
deduce origin
By Michael Schroeder, Biotec, 2004
131
Example:
Westnil virus in NY
How can the trees be
constructed?
By Michael Schroeder, Biotec, 2004
132
Three Methods to Generate
Phylogenetic Trees
Distance-based
Hierarchical clustering
Character-based
Parsimony
Maximum likelihood
By Michael Schroeder, Biotec, 2004
133
Distance-based Approach
Single Alignment
atgctctggccacggcacttgcggatcccagggtgatctgtgcacctgcgata
||||||||||||||| |||| ||||||||
|||| |||||||||||||||
atgctctggccacggatcttgtggatccca---tgatatgtgcacctgcgata
Score: 46 matches, 3 mismatches, 1 gap, 3 gap extensions,
z.B. Score = 46x1 - 3x1 - 1x2 - 3x1 = 38
Approach:
Define distance between two sequences, e.g. percentage of
mismatches in their alignment
Construct tree, which groups sequences with minimal
distances iteratively together
By Michael Schroeder, Biotec, 2004
134
1
1
2
3
4
5
0
2
6
10
9
0
5
9
8
0
4
5
0
3
2
3
4
5
(1,2)
Hierarchical Clustering
0
(1,2) 3
4
5
0
5
9
8
0
4
5
4
0
3
3
3
4
5
5
0
(1,2)
3
(4,5)
0
5
8
1
0
4
0
3
(4,5)
0
(1,2)
(3,(4,5))
2
(1,2)
(1,2)
(3,(4,5))
0
5
1
2
3
4
5
0
By Michael Schroeder, Biotec, 2004
135
Hierarchical Clustering
Given a distance matrix D=(dij) with 1≤ i,j ≤ n
Result: A binary tree of clusters
Init:
ToDo = {}
For all i in { 1,…, n } do
Let ti be a tree without children, i.e. a leaf
ToDo := ToDo { ti }
Main loop
While |ToDo | > 1 do
Find i,j such that dij is minimal
Add a new column and row labelled k := (i,j) to D
For all indices h of D apart from k,i,j do
dh,k = dk,h := min { dh,i , dh,j }
// min = single linkage
Let tk be a new tree with children ti and tj
ToDo := ( ToDo { tk } ) - { ti ,tj }
Remove columns and rows i,j from D
Complexity: O(n2)
By Michael Schroeder, Biotec, 2004
136
Hierarchical Clustering:
How to define distance between clusters?
Single linkage:
dh,k = dk,h := min { dh,i , dh,j }
Example: Distance (A,B) to C is 1
A
B
Complete linkage:
C
A
B
C
0
1
2
0
1
0
dh,k = dk,h := max { dh,i , dh,j }
Example: Distance (A,B) is C is 2
Average linkage:
dh,k = dk,h := 0.5 dh,i + 0.5 dh,j
Example: Distance (A,B) to C is 1.5
Are dendrograms always the same
independent of the linkage method?
By Michael Schroeder, Biotec, 2004
A B C
A B C
137
Parsimony-method
Approach: Generate “smallest” tree
containing all the sequences as leaves
Seq
a
b
c
d
1
G
G
G
G
2
G
G
G
A
3
G
G
A
T
4
G
A
T
C
5
G
G
A
A
a GGGGGG
6
G
T
G
T
3 G->A
4 G->T
5 G->A
4 G->A
6 G->T
b GGGAGT c GGATAG
By Michael Schroeder, Biotec, 2004
2 G->A
3 T->A
4 T->C
6 G->T
d GATCAT
138
Parsimony
Generate smallest tree
Informative vs. non-informative sites
Build pairs with fewest possible substitutions
Example:
3 possible trees:
((a,b),(c,d)) or ((a,c),(b,d)) or ((a,d),(b,c))
1,2,3,4 are not informative
5,6 are informative
5: ((a,b),(c,d))
6: ((a,c),(b,d))
By Michael Schroeder, Biotec, 2004
Seq
a
b
c
d
1
G
G
G
G
2
G
G
G
A
3
G
G
A
T
4
G
A
T
C
5
G
G
A
A
6
G
T
G
T
139
Maximum likelihood
Assigns quantitative probabilities to mutation
events
Reconstructs ancestors for all nodes in the tree
Assigns branch lengths based on probabilities of the
mutational events
For each possible tree topology, the assumed
substitution rates are varied to find the parameters
that give the highest likelihood of producing the
observed data
By Michael Schroeder, Biotec, 2004
140
Problems
Character-based methods tend to be better (based
on paleontological data)
All make assumptions:
No back mutations
Same evolutionary rate
By Michael Schroeder, Biotec, 2004
141
Assessing Quality: Bootstrapping
Given a tree obtained from one of the methods above
Generate Multiple Alignment
For a number of interations
Generate new sequences by selecting columns
(possibly the same column more than once) form
the multiple alignment
Generate tree for the new sequences
Compare this new tree with the given tree
For each cluster in the given tree, which also approach
in the new tree, the bootstrap value is increased
Bootstrap-Value = Percentage of trees containing the
same cluster
By Michael Schroeder, Biotec, 2004
142
From where are we?
Recent-Africa Hypothesis
Homo Sapiens came 100-200.000 years ago from
Africa
Multi-regional Hypothesis
Ancestors of Homo Sapiens left Africa ca. 2.000.000
years ago
Which one’s right?
By Michael Schroeder, Biotec, 2004
143
Experiment
Mitochondrial DNA form 53
humans in different regions
sequenced
Outgroup = Mitochondrial
DNA of chimpanzee
By Michael Schroeder, Biotec, 2004
144
A nice phylogeny (Nature 2004)
Nature October 2004 Volume 431 No. 7012
By Michael Schroeder, Biotec, 2004
145
Why Mitochondria?
Simple genetic structure
No repetitions
No Pseudo genes
No Introns
No recombination
By Michael Schroeder, Biotec, 2004
146
Molecular Clock
Based on genetic and paleontological the most recent common
ancestor (mrca) of chimpanzee and homo sapiens dates back
5.000.000 years
Molecular clock:
1.7 x10-8 nucleotide changes per site and year
Assumption: equal distribution, no silent mutations
Diversity in Afrikca: 3.7 x10-3 nucleotide changes per site and year
Diversity outside Africa: 1.7 x10-3 nucleotide changes per site and year
Estimated expansion1925 generations ago = ca. 40.000 years
Mrca of all humans: 171.500 +/- 50.000 years ago
Mrca of African and non-African: 52,000 +/- 27.500 years ago
Experiment supports recent-Africa hypothesis
By Michael Schroeder, Biotec, 2004
147
Summary
Schools of taxonomists
Assumptions made
Methods
Distance-based
Character-based
By Michael Schroeder, Biotec, 2004
148