Macromolecular sequence analysis

PAM and BLOSUM

substitution matrices

Didier Gonze - 20/9/2015

Substitution scoring matrices

There are two main families of amino acids substitution
scoring matrices:
PAM substitution matrices
based on the rate of divergence between sequences
BLOSUM substitution matrices
based on the conservation of domains in proteins

Another popular substitution matrix was proposed by

Gonnet et al (1992):
GONNET substitution matrix
based on an exhaustive sequence alignment analysis

PAM scoring matrices

The substitution score is expected to depend
on the rate of divergence between sequences.
accepted mutations

The PAM matrices derived by Dayhoff (1978):

are based on evolutionary distances.
have been obtained from carefully aligned closely
related protein sequences (71 gapless alignments of
sequences having at least 85% similarity).
M. Dayhoff

Dayhoff et al. (1978). A model of evolutionary change in proteins. In Atlas of Protein

Sequence and Structure, vol. 5, suppl. 3, 345352. National Biomedical Research
Foundation, Silver Spring, MD, 1978.

PAM scoring matrices

PAM = Percent (or Point) Accepted Mutation
The PAM matrices are series of scoring matrices, each reflecting a
certain level of divergence:
PAM = unit of evolution (1 PAM = 1 mutation/100 amino acid)
PAM1
PAM50
PAM250

proteins with an evolutionary distance of 1% mutation/position

idem for 50% mutations/position
250% mutations/position (a position could mutate several times)

Dayhoff et al. (1978). A model of evolutionary change in proteins. In Atlas of Protein

Sequence and Structure, vol. 5, suppl. 3, 345352. National Biomedical Research
Foundation, Silver Spring, MD, 1978.

Derivation of the PAM matrices

To illustrate how the PAM substitution matrices have been derived, we
will consider the following artificial ungapped aligned sequences:

A
D
A
C

C
B
D
B

G
G
I
I

H
H
J
J

Example taken from Borodovsky & Ekisheva (2007) Problems and Solutions in Biological
sequence analysis. Cambridge Univ Press.

Derivation of the PAM matrices

Phylogenetic trees (maximum parsimony)

ABGH

ABIJ
H-J G-I

J-H I-G

ABIJ

ABGH

ABGH
B-C

A-D

ACGH

B-D

DBGH

A-C

ADIJ

CBIJ

B-C

ABIJ
A-D

ACGH

DBGH

ABIH
I-G

ACGH

ABIJ

DBGH

ADIJ

J-H

H-J

A-D

A-C

CBIJ

ABGJ

ABGH
B-C

B-D

B-D

ADIJ

G-I

ABGH

A-C

CBIJ

B-C

ACGH

ABIJ
A-D

DBGH

B-D

ADIJ

A-C

CBIJ

Here are represented the four more parsimonious (minimum of substitutions)

phylogenetic trees for the alignment given above.

Derivation of the PAM matrices

Matrix of accepted point mutation counts (A)
A
A

0
0

For each pair of different

amino acids (i,j), the total
number aij of substitutions
from i to j along the edges of
the phylogenetic tree is
calculated.
(they are indicated in blue on the
previous slide)

Derivation of the PAM matrices

Each edge of a given tree is associated with the ungapped
alignment of the two sequences connected by this edge.
Thus, any tree shown above generates 6 alignments. For example
the first phylogenetic tree generates the following alignments:

ABGH
H-J G-I

ABGH
B-C

ABIJ
A-D

ACGH

DBGH

B-D

ADIJ

A-C

CBIJ

A B G H
A B G H

A B G H
A B I J

A B G H
A C G H

A B G H
D B G H

A B I J
A D I J

A B I J
C B I J

Those alignments can be used to assess the "relative mutability"

of each amino acid.

Derivation of the PAM matrices

Relative mutability (mi)
The relative mutability is defined by the ratio of the total number of
times that amino acid j has changed in all the pair-wise alignments
(in our case 6x4=24 alignments) to the number of times that j has
occurred in these alignments, i.e.

number of changes of j
mj =
number of occurrences of j
Relative amino acid mutability values mj for our example
Amino acid

Changes (substitutions)

Frequency of occurrence

40

40

24

24

24

24

Relative mutability mj

0.2

0.2

0.167

0.167

0.167

0.167

The relative mutability accounts for the fact that the different amino acids have different
mutation rates. This is thus the probability to mutate.

Derivation of the PAM matrices

Relative mutability of the 20 amino acids
aa

mi

aa

mi

Asn

134

His

66

Ser

120

Arg

65

Asp

106

Lys

56

Glu

102

Pro

56

Ala

100

Gly

49

Thr

97

Tyr

41

Ile

96

Phe

41

Met

94

Leu

40

Gln

93

Cys

20

Val

74

Trp

18

Values according to Dayhoff (1978)

The value for Ala has been arbitrarily
set at 100.

Trp and Cys are less mutable

Cys is known to have several unique, indispensable
function (attachment site of heme group in cytochrome
and of F eS clusters in ferredoxin). It also forms cross-links
such as in chymotrypsin or ribonuclease.
Big groups like Trp or Phe are less mutable due to their
particular chemistry. On the other extreme, the low
mutability of Cys must be due to its unique smallness that
is advatageous in many places.

Asn, Ser, Asp and Glu are most mutable

Although Ser sometimes functions in the active center, it
more often performs a function of lesser importance, easily
mimicked by several other amino acids of similar physical
and chemical properties.

Derivation of the PAM matrices

Effective frequency (fi)
The notion of effective frequency fi takes into account the difference in variability
of the primary structure conservation in proteins with different functional roles.
Two alignment blocks corresponding to 2 different families may contribute
differently to fi even if the number of occurrence of amino acid j in these blocks is
the same.

" relative frequency of %

" average composition% " number of mutations in%
$
' = $
' $
'
# exposure to mutation &
# of each group & # the corresponding tree &

Derivation of the PAM matrices

Effective frequency (fi)
The effective frequency is defined as

f j = k q(bj )N (b )
b

where

the sum is taken over all alignment blocks b

qj(b) is the observed frequency of amino acid j in block b,
N(b) is the number of substitutions in a tree built for b

and the coefficient k is chosen the ensure that the sum of the frequences fj = 1.

In our example, there is only one block, therefore the effective frequencies
are equal to the compositional frequencies (fi = qj):
Amino acid

Frequency f

0.125

0.125

0.125

0.125

0.125

0.125

0.125

0.125

Derivation of the PAM matrices

Effective frequency of the 20 amino acids determined
for the original alignment data (Dayhoff et al., 1978)
Amino acid

Gly

Ala

Leu

Lys

Ser

Val

Thr

Frequency f

0.089

0.087

0.085

0.081

0.070

0.065

0.058

Amino acid

Pro

Glu

Asp

Arg

Asn

Phe

Gln

Frequency f

0.051

0.050

0.047

0.041

0.040

0.040

0.038

Amino acid

Ile

His

Cys

Tyr

Met

Trp

Frequency f

0.037

0.034

0.033

0.030

0.015

0.010

Source: Dayhoff, 1978

Distribution of amino acids found in 1081

peptides and proteins listed in the Atlas of
Protein Sequence and Structure (1981).
Doolittle RF (1981) Similar amino acid
sequences: c hance or common ancestry?
Science. 214:149-59.

Derivation of the PAM matrices

Mutational probability matrix (M)
Let's define Mij the probability of the amino acid in column j having been
substituted by an amino acid in row i over a given evolutionary time unit.
Non-diagonal elements of M:

M ij =

m j Aij

Diagonal elements of M:

M ii = 1 m i

Akj
k

In these equations, m is the relative mutability and A is the matrix of accepted point mutations.
The constant represents a degree of freedom that could
be used to connect the matrix M
with an evolutionary
time scale.

In our example:
A
A

0
4
4

see matrix A

this represents
32/40 of the cases

B
C
D

this represents
8/40 of the cases
mutability m

If A is mutated, the probability that

it is mutated into D is
ADA/(ABA+ACA+ADA) = 4/8
Thus the probability that A is
mutated into D is:
MDA = 4/8 * 8/40 = 4/40
and the probability that A is not
mutated is:
MAA = 1 - 8/40 = 32/40

Derivation of the PAM matrices

Mutational probability matrix (M)
Let's define Mij the probability of the amino acid in column j having been
substituted by an amino acid in row i over a given evolutionary time unit.
Non-diagonal elements of M:

M ij =

Diagonal elements of M:

m j Aij

M ii = 1 m i

Akj
k

In these equations, m is the relative mutability and A is the matrix of accepted point mutations.
The constant represents a degree of freedom that could
be used to connect the matrix M
with an evolutionary
time scale.

The coefficient could be adjusted to ensure that a specific (small) number of substitutions would
occur on average per hundred residues. This adjustement was done by Dayhoff et al in the following
way. The expected number of amino acids that will remain inchanged in a protein sequence 100
amino acid long is given by:

100 f j M jj = 100 f j (1 m j )
j

If only one substitution per residue is allowed, then is calculated from the equation:

100 f j (1 m j ) = 99
j

Derivation of the PAM matrices

Mutational probability matrix
In our example, = 0.0261 and the mutation probability matrix (PAM1) is:

0.9948

0.0131

0.0131

0.9948

0.0131

0.0131

0.0026

0.0026

0.9740

0.0026

0.0026

0.9740

0.9957

0.0043

0.9957

0.0043

0.0043

0.9957

0.0043

0.9957

Note that M is a non-symmetric matrix.

Derivation of the PAM matrices

Mutational probability matrix derived by Dayhoff for the 20 amino acids
A

9867

10

17

21

22

35

32

18

R
N
D
C
Q
E
G
H
I
L
K
M
F
P
S
T
W
Y
V

9913

10

10

19

9822

36

21

13

20

42

9859

53

9973

9876

27

23

10

56

35

9865

21

12

11

9935

21

18

20

9912

9872

12

33

22

9947

45

13

15

37

25

12

9926

20

11

9874

9946

28

13

9926

12

28

11

34

11

16

17

9840

38

22

13

11

32

9871

9976

21

9945

13

57

11

17

10

9901

For clarity, the values have been multiplied by 10000

This matrix corresponds to an evolution time period giving 1
mutation/100 amino acids, and is refered to as the PAM1 matrix.

Source: Dayhoff, 1978

Derivation of the PAM matrices

Mutational probability matrix derived by Dayhoff for the 20 amino acids
This matrix is the mutation
probability matrix for an evolution
time of 1 PAM.
The diagonal represents the
probability to still observe the same
residue after 1 PAM. Therefore the
diagonal represents the 99% of the
case of non-mutation.
Note that this does not mean that there
was no mutation during this time interval.
Indeed, the conservation of a residue
could reflect either a conservation during
the whole period, or a succession of two
or more mutations ending at the initial
residue
Source: J. v an Helden

Derivation of the PAM matrices

From PAM1 to PAM2

line 3 of PAM1

column 17 of PAM1

=> Matrix product: PAM2 = PAM1 x PAM1

Source: J. v an Helden

Derivation of the PAM matrices

From PAM1 to PAM2, PAM100, PAM250, etc...
Remark (from graph theory)

Source: J. v an Helden

...

...

...

...

...

...

...

...

...

...

...

...

...

...

...

...

Matrix Q indicates the

number of paths going from
one node to another in 1
step

Matrix Q2 indicates the

number of paths going from
one node to another in 2
steps

Matrix Qn indicates the

number of paths going from
one node to another in n
steps

Derivation of the PAM matrices

From PAM1 to PAM2, PAM100, PAM250, etc...
Similarly:
PAM1
PAM2 = PAM12
PAM100 = PAM1100
PAM250 = PAM1250
PAMn = PAM1n

gives the probability to observe the changes i j per 100 mutations

gives the probability to observe the changes i j per 200 mutations
gives the probability to observe the changes i j per 10 000 mutations
gives the probability to observe the changes i j per 25 000 mutations
gives the probability to observe the changes i j per 100xn mutations.

Convergence: it can be verified that

$ fA
&
fR
n
&

PAM = PAM1 converges to the observed frequencies: lim M =

n
& ...
&
% fV
Dayhoff et al. (1978) checked this convergence by computing M2034.

fA
fR
...

...
...

fV

...

fA '
)
fR )
... )
)
fV (

Derivation of the PAM matrices

PAM250 derived by Dayhoff for the 20 amino acids
A
R
N
D
C
Q
E
G
H
I
L
K
M
F
P
S
T
W
Y
V

13
3
4
5
2
3
5
12
2
3
6
6
1
2
7
9
8
0
1
7

6
17
4
4
1
5
4
5
5
2
4
18
1
1
5
6
5
2
1
4

9
4
6
8
1
5
7
10
5
2
4
10
1
2
5
8
6
0
2
4

9
3
7
11
1
6
11
10
4
2
3
8
1
1
4
7
6
0
1
4

5
2
2
1
52
1
1
4
2
2
2
2
0
1
3
7
4
0
3
4

8
5
5
7
1
10
9
7
7
2
6
10
1
1
5
6
5
0
1
4

9
3
6
10
1
7
12
9
4
2
4
8
1
1
4
7
5
0
1
4

12
2
4
5
2
3
5
27
2
2
3
5
1
1
5
9
6
0
1
4

6
6
6
6
2
7
6
5
15
2
5
8
1
3
5
6
4
1
3
5

8
3
3
3
2
2
3
5
2
10
15
5
2
5
3
5
6
0
2
4

6
2
2
2
1
3
2
4
2
6
34
4
3
6
3
4
4
1
2
15

7
9
5
5
1
5
5
6
3
2
4
24
2
1
4
7
6
0
1
10

7
4
3
3
1
3
3
5
2
6
20
9
6
4
3
5
5
0
2
4

4
1
2
1
1
1
1
3
2
5
13
2
2
32
2
3
3
1
15
10

11
4
4
4
2
4
4
8
3
2
5
6
1
1
20
9
6
0
1
5

11
4
5
5
3
3
5
11
3
3
4
8
1
2
6
10
8
1
2
5

11
3
4
5
2
3
5
9
2
4
6
8
1
2
5
9
11
0
2
5

2
7
2
1
1
1
1
2
2
1
6
4
1
4
1
4
2
55
3
72

4
2
3
2
4
2
2
3
3
3
7
3
1
20
2
4
3
1
31
4

9
2
3
3
2
3
3
7
2
9
13
5
2
3
4
6
6
0
2
17

For clarity, the values have been multiplied by 100

This matrix corresponds to an evolution time period giving 250
mutation/100 amino acids (i.e. an evolutionary distance of 250 PAM),
and is refered to as the PAM250 matrix.

Source: Dayhoff, 1978

Derivation of the PAM matrices

Interpretation of the PAM250 matrix
In comparing 2 sequences at this evolutionary distance
(250 PAM), there is:

* * * * A * * * * *
250 PAM
*
*
*
*
...

*
*
*
*

*
*
*
*

*
*
*
*

A
R
N
W

*
*
*
*

*
*
*
*

*
*
*
*

*
*
*
*

*
*
*
*

probability o f 13%
probability o f 3%
probability o f 4%
probability o f 0%

Source: Dayhoff, 1978

Derivation of the PAM matrices

From probabilities to scores
So far, we have obtained a probability matrix, but we would like a scoring
matrix.
A score should reflect the significance of an alignment occurring as a result of
an evolutionary process with respect to what we could expect by chance.
A score should involve the ratio between the probability derived from non-
random (evolutionary) to random models:

M nji P ji,n
rn (i, j) =
=
fj
fi f j

probability to see a pair (i,j) due to evolution

probability to see a pair (i,j) by chance

The matrix Mjin is the mutational probability matrices at PAM distance n.

Matrices M1 and M250 have been shown before.

Pji,n = fi Mjin is the probability that two aligned amino acids have diverged
from a common ancestor n/2 PAM unit ago, assuming that the substitutions
follow a Markov process (for details, see Borodovsky & Ekisheva, 2007).
Note that R (the odd-score or relatedness matrix) is a symmetric matrix.

Derivation of the PAM matrices

Log-odd scores
In practice, we often use the log-odd scores defined by

sn (i, j) = log

M nji
fj

= log

P ji,n
fi f j

This definition has convenient practical consequences:

A positive score (sn > 0) characterizes the accepted mutations

A negative score (sn < 0) characterizes the unfavourable mutations

Another property of the log-odd scores is that they can be added to

produce the score of an alignment:
T
Y

A
S

H
D

G
G

K
D

Salignment = s(T,Y) + s(A,S) + s(H,D) + s(G,G) + s(K,D)

Derivation of the PAM matrices

PAM250 matrix: log-odds scores
A

-2

-2

-1

-2

-1

-1

-3

-6

-4

R
N
D
C
Q
E
G
H
I
L
K
M
F
P
S
T
W
Y
V

-2
0
0
-2
0
0
1
-1
-1
-2
-1
-1
-3
1
2
0
-6
-3
0

6
0
-1
-3
1
-1
-3
1
-2
-3
3
-1
-4
0
1
-2
2
-5
-2

0
2
2
-4
1
1
0
1
-2
-3
1
-2
-3
0
2
0
-5
-2
-2

-1
2
4
-5
2
3
1
1
-2
-4
0
-3
-5
-1
1
-1
-7
-4
-2

-4
-3
-5
12
-5
-5
-3
-3
-2
-6
-5
-5
-4
-3
1
-3
-7
1
-2

1
1
2
-5
4
2
-1
3
-2
-2
1
-1
-4
0
0
-2
-6
-4
-2

-1
2
3
-5
2
4
0
0
-2
-3
0
-2
-5
0
1
-1
-7
-4
-2

-3
1
1
-4
-1
0
5
-3
-3
-4
-2
-3
-4
0
2
-1
-7
-5
-2

2
2
1
-3
3
1
-2
6
-3
-2
0
-2
-2
0
0
-2
-5
0
-2

-2
-2
-2
-3
-2
-2
-2
-3
4
2
-2
2
1
-2
-1
-1
-6
-1
4

-3
-3
-4
-6
-2
-3
-4
-3
2
6
-3
4
2
-2
-2
-2
-7
-1
2

3
1
0
-5
1
0
-2
0
-2
-2
5
1
-5
-1
1
-1
-4
-5
-2

0
-2
-3
-5
-1
-2
-3
-3
2
4
0
6
0
-2
-1
-1
-6
-2
2

-4
-3
-5
-4
-4
-5
-5
-2
1
2
-5
0
9
-4
-2
-4
1
7
-1

0
0
-1
-2
0
0
0
0
-2
-2
-1
-2
-5
6
2
0
-6
-5
-1

0
1
0
0
-1
0
1
-1
-1
-3
0
-2
-3
1
2
1
-2
-3
-1

-1
1
0
-2
-1
0
0
-1
0
-2
0
0
-3
0
2
2
-5
-3
0

2
-4
-7
-8
-5
-7
-7
-3
-5
-2
-4
-4
0
-6
-2
-6
17
0
-6

-5
-2
-4
0
-4
-4
-5
0
-1
-1
-5
-3
7
-5
-2
-4
1
10
-3

-2
-2
-2
-2
-2
-2
-1
-3
4
2
-2
2
-1
-1
0
0
-8
-2
4

For clarity, the values have been multiplied by 10

Source: Dayhoff, 1978

PAM scoring matrices

Hinton diagram
Yellow boxes indicate
positive values
(accepted mutations).
Red box indicate
negative values
(avoided mutations).
The aera of each box
is proportional to the
absolute value of the
log-odds scores.

Source: J. v an Helden

Derivation of the PAM matrices

Cys
Ser
Thr
Pro
Ala
Gly
Asn
Asp
Glu
Gln
His
Arg
Lys
Met
Ile
Leu
Val
Phe
Tyr
Trp

C
S
T
P
A
G
N
D
E
Q
H
R
K
M
I
L
V
F
Y
W

12
PAM250 matrix (log-odds)
0
2
-2
1
3
-1
1
0
6
-2
1
1
1
2
-3
1
0
-1
1
5
-4
1
0
-1
0
0
2
-5
0
0
-1
0
1
2
4
-5
0
0
-1
0
0
1
3
4
-5
-1
-1
0
0
-1
1
2
2
4
-3
-1
-1
0
-1
-2
2
1
1
3
6
-4
0
-1
0
-2
-3
0
-1
-1
1
2
6
-5
0
0
-1
-1
-2
1
0
0
1
0
3
5
-5
-2
-1
-2
-1
-3
-2
-3
-2
-1
-2
0
0
6
-2
-1
0
-2
-1
-3
-2
-2
-2
-2
-2
-2
-2
2
5
-6
-3
-2
-3
-2
-4
-3
-4
-3
-2
-2
-3
-3
4
2
6
-2
-1
0
-1
0
-1
-2
-2
-2
-2
-2
-2
-2
2
4
2
4
-4
-3
-3
-5
-4
-5
-4
-6
-5
-5
-2
-4
-5
0
1
2
-1
9
0
-3
-3
-5
-3
-5
-2
-4
-4
-4
0
-4
-4
-2
-1
-1
-2
7 10
-8
-2
-5
-6
-6
-7
-4
-7
-7
-5
-3
2
-3
-4
-5
-2
-6
0
0
C
S
T
P
A
G
N
D
E
Q
H
R
K
M
I
L
V
F
Y
Cys Ser Thr Pro Ala Gly Asn Asp Glu Gln His Arg Lys Met Ile Leu Val Phe Tyr
Hydrophobic
C
P
A
G
M
I
L
V
Aromatic
H
F
Y
Polar
S
T
N
Q
Y
Basic
H
R
K
Acidic
D
E

Source: J. v an Helden

17
W
Trp
W

PAM matrices: exercise

The original PAM250 substitution matrix scores a substitution of Gly by Arg
by a negative score -3 (decimal logarithm and scaling factor 10 are used,
with rounding to the nearest neighbour). The average frequency of Arg in
the protein sequence database is 0.041. Use this information as well as the
method described above to estimate the probability that Gly will be
substituted by Arg after a PAM250 time period.

Source: Borodovsky & Ekisheva (2007)

PAM matrices: exercise

The original PAM250 substitution matrix scores a substitution of Gly by Arg
by a negative score -3 (decimal logarithm and scaling factor 10 are used,
with rounding to the nearest neighbour). The average frequency of Arg in
the protein sequence database is 0.041. Use this information as well as the
method described above to estimate the probability that Gly will be
substituted by Arg after a PAM250 time period.
The element sij of the PAM250 substitution matrix and the frequency of
amino acid j (fj) in a protein sequence database are connected by the
following formula:
#
P(i j in 250 PAM) &
sij = %%10log
((
fj
$
'
Therefore, the probability of substitution of Gly by Arg is:

P(Gly Arg in 250 PAM) = 0.041100.3 = 0.0205

Source: Borodovsky & Ekisheva (2007)

Derivation of the PAM matrices

Scoring an alignment
A scoring matrix like PAM250 can be
used to score an alignment

T
Y

A
S

H
D

G
G

K
D

Salignment = s(T,Y) + s(A,S) + s(H,D) + s(G,G) + s(K,D)

= -3 + 1 + 1 + 5 + 0
= 4

Choosing the appropriate PAM matrix

How to choose the appropriate PAM matrix?
Correspondance between the observed percent of amino acid difference d
between the evolutionary distance n (in PAM) between them:

100 f j M njj = 100 d

j

twilight zone
(detection limit)

identity
(%)

difference
d (%)

PAM index
n

99

95

90

10

11

85

15

17

80

20

23

75

25

30

70

30

38

60

40

56

50

50

80

40

60

112

30

70

159

20

80

246

14

86

350

Choosing the appropriate PAM matrix

How to choose the appropriate PAM matrix?
Altschul SF(1991) Amino acid substitution matrices from an
information theoretic perspective. J Mol Biol. 219:555-65.
PAM120 matrix is the most appropriate for database searches
PAM200 matrix is the most appropriate for comparing two specific
proteins with suspected homology

Remark:
In the PAM matrices, the index indicates the percentage of
substitution per position.
Higher indexes are more appropriate for more distant proteins
(PAM250 better than PAM100 for distant proteins).

Improved PAM matrices

Update of PAM matrices
With the rapid growth of protein data, updates and variants of
the PAM matrices series have been proposed:
Jones DT, Taylor WR, Thornton JM (1992) The rapid generation of mutation
data matrices from protein sequences. Comput Appl Biosci. 8: 275-82.
Whelan S, Goldman N (2001) A general empirical model of protein evolution
derived from multiple protein families using a maximum-likelihood approach.
Mol Biol Evol. 18 : 691-9.

BLOSUM scoring matrices

BLOSUM matrices were designed to find conserved regions in
proteins (Henikoff & Henikoff, 1992).
Contrarily to the PAM matrices, the BLOSUM matrices:
are not based on evolutionary distances
are based on ungapped aligned regions from proteins families
called the block database (Henikoff & Henikoff, 1991)

BLOSUM = BLOcks SUbstitution Matrix

Henikoff S and Henikoff JG (1992). Amino acid substitution matrices from
protein blocks. PNAS 89:10915-10919.

BLOSUM scoring matrices

Collection of protein blocks
clustering

BLOSUM62

62%
MOTIF
80%

clusters of sequences
sharing a high
identity

...
504 groups of related
proteins

BLOSUM80

2205 blocks (short

domains of ungapped
multiple alignment)

Henikoff S and Henikoff JG (1991). Automated assembly of protein blocks for

database searching. Nucl Acids Res 23:6565-6572.
Web: http://blocks.fhcrc.org/blocks/

Derivation of the BLOSUM matrices

As done for the PAM matrices, we will illustrate how the
BLOSUM matrices can be derived using a simple example.
Let's start from the following sample sequences:
A D A D
A D C D
A C C D
D C A A
D C A A
A A C C
D A C C

The sequences are first

clustered into "blocks"
according the % of identity.
(here: 75% - the BLOSUM
matrix obtained will thus have
the index 75)

Example taken from Borodovsky & Ekisheva

(2007) Problems and Solutions in Biological
sequence analysis. Cambridge Univ Press.

Derivation of the BLOSUM matrices

Matrix of weighted counts (F)
A D A D
A D C D
A C C D

each element in the first block

has a weight of 1/3

D C A A
D C A A

each element in the second block

has a weight of 1/2

A A C C
D A C C

each element in the third block

has a weight of 1/2

Derivation of the BLOSUM matrices

Matrix of weighted counts (F)
1/3
1/3
1/3

1/2

A D A D
A D C D
A C C D

weight:
1/3

D C A A
D C A A

weight:
1/2

A A C C
D A C C

weight:
1/2

5/6

13/3

11/3

13/3

5/3

11/3

5/3

1/2

Each element fij of matrix F is the weighted

count of substitution of element i (in a cluster)
by j (in another cluster).
For example, here
fAA = 1/3*1/2 + 1/3*1/2 + 1/3*1/2 + 0 + 1/3*1/2 + 1/3*1/2 + 0 = 5/6
1st
column

2nd
column

3rd
column

4th
column

Derivation of the BLOSUM matrices

Observed frequencies of occurrence (Q)
The observed frequencies of occurrence of a pair (i,j) is defined by:
f ij

qij =

f
i

ij

j=1

In our example, we get the following observed frequencies of occurrence:

fij

qij

5/6

13/3

11/3

5/72

13/36

11/36

5/3

1/12

5/36

1/2

C
D
i

f
i

j=1

ij

= fAA+fAC+fAD+fCC+fCD+fDD = 12

Examples:

1/24
qAA = fAA/12 = 5/72
qCA = fCA/12 = fAC/12 = 5/72

Derivation of the BLOSUM matrices

Expected frequency (E)
By definition, the expected frequencies are:

eii = pi2

eij = 2 pi p j (for ij)

where pi is the probability of occurrence of amino acid i:

1
pi = qii + qij
2 ji

we assume that there are as

much C A than A C
(qij = pij + pji)

In our example, we get the following expected frequencies:

pA = qAA + 1/2 (qCA) + 1/2 (qDA) = 29/72

pC = qCC + 1/2 (qAC) + 1/2 (qDC) = 19/72

pD = qDD + 1/2 (qAD) + 1/2 (qCD) = 1/3

eij

0.1622

0.2683

0.2125

0.1108

0.1757

C
D

0.0696

Derivation of the BLOSUM matrices

Log-odd ratio (S)
Finally, we calculate the log-odd ratio as
Observed frequencies of
occurrence of a pair (i,j)

qij
sij = 2log 2
eij

Expected frequencies of
occurrence of a pair (i,j)

Calculating the log-odd ratio in our example, we obtained the

following substitution score matrix:

s ij

-2

-1

-1

-1

-1

This matrix corresponds to the BLOSUM matrix

Because the original clustering was done for a threshold identity of
75%, this is a BLOSUM75 matrix

5
0
-2
-3
1
0
-2
0
-3
-2
2
-1
-3
-2
-1
-1
-3
-2
-3
-1
0
-1
-4

6
1
-3
0
0
0
1
-3
-3
0
-2
-3
-2
1
0
-4
-2
-3
3
0
-1
-4

6
-3
0
2
-1
-1
-3
-4
-1
-3
-3
-1
0
-1
-4
-3
-3
4
1
-1
-4

9
-3
-4
-3
-3
-1
-1
-3
-1
-2
-3
-1
-1
-2
-2
-1
-3
-3
-2
-4

5
2
-2
0
-3
-2
1
0
-3
-1
0
-1
-2
-1
-2
0
3
-1
-4

5
-2
0
-3
-3
1
-2
-3
-1
0
-1
-3
-2
-2
1
4
-1
-4

6
-2
-4
-4
-2
-3
-3
-2
0
-2
-2
-3
-3
-1
-2
-1
-4

8
-3
-3
-1
-2
-1
-2
-1
-2
-2
2
-3
0
0
-1
-4

4
2
-3
1
0
-3
-2
-1
-3
-1
3
-3
-3
-1
-4

4
-2
2
0
-3
-2
-1
-2
-1
1
-4
-3
-1
-4

5
-1
-3
-1
0
-1
-3
-2
-2
0
1
-1
-4

5
0
-2
-1
-1
-1
-1
1
-3
-1
-1
-4

6
-4
-2
-2
1
3
-1
-3
-3
-1
-4

7
-1
-1
-4
-3
-2
-2
-1
-2
-4

4
1
-3
-2
-2
0
0
0
-4

5
-2
-2
0
-1
-1
0
-4

11
2
-3
-4
-3
-2
-4

7
-1
-3
-2
-1
-4

4
-3 4
-2 1 4
-1 -1 -1 -1
-4 -4 -4 -4

Asp

Cys

Gln

Glu

Gly

His

Ile

Leu

Lys

Met

Phe

Pro

Ser

A
A

C
C

G
G

I
I

L
L

M
M

P
P

S
H

V
V

Y
Y

Val

Tyr

Trp

Here is the BLOSUM matrix obtained by Henikoff

& Henikoff on the basis of 2205 block of proteins.
In each group, the sequences have been
clustered together if they were 62% identical.
This matrix is referred to as BLOSUM62.

Thr

Hydrophobic
Aromatic
Polar
Basic
Acidic

4
-1
-2
-2
0
-1
-1
0
-2
-1
-1
-1
-1
-2
-1
1
0
-3
-2
0
-2
-1
0
-4

Asn

A
R
N
D
C
Q
E
G
H
I
L
K
M
F
P
S
T
W
Y
V
B
Z
X
*

Arg

Ala
Arg
Asn
Asp
Cys
Gln
Glu
Gly
His
Ile
Leu
Lys
Met
Phe
Pro
Ser
Thr
Trp
Tyr
Val

Ala

BLOSUM62 matrix

Source: J. v an Helden

5
0
-2
-3
1
0
-2
0
-3
-2
2
-1
-3
-2
-1
-1
-3
-2
-3
-1
0
-1
-4

6
1
-3
0
0
0
1
-3
-3
0
-2
-3
-2
1
0
-4
-2
-3
3
0
-1
-4

6
-3
0
2
-1
-1
-3
-4
-1
-3
-3
-1
0
-1
-4
-3
-3
4
1
-1
-4

9
-3
-4
-3
-3
-1
-1
-3
-1
-2
-3
-1
-1
-2
-2
-1
-3
-3
-2
-4

5
2
-2
0
-3
-2
1
0
-3
-1
0
-1
-2
-1
-2
0
3
-1
-4

5
-2
0
-3
-3
1
-2
-3
-1
0
-1
-3
-2
-2
1
4
-1
-4

6
-2
-4
-4
-2
-3
-3
-2
0
-2
-2
-3
-3
-1
-2
-1
-4

8
-3
-3
-1
-2
-1
-2
-1
-2
-2
2
-3
0
0
-1
-4

4
2
-3
1
0
-3
-2
-1
-3
-1
3
-3
-3
-1
-4

4
-2
2
0
-3
-2
-1
-2
-1
1
-4
-3
-1
-4

5
-1
-3
-1
0
-1
-3
-2
-2
0
1
-1
-4

5
0
-2
-1
-1
-1
-1
1
-3
-1
-1
-4

6
-4
-2
-2
1
3
-1
-3
-3
-1
-4

7
-1
-1
-4
-3
-2
-2
-1
-2
-4

4
1
-3
-2
-2
0
0
0
-4

5
-2
-2
0
-1
-1
0
-4

11
2
-3
-4
-3
-2
-4

7
-1
-3
-2
-1
-4

4
-3 4
-2 1 4
-1 -1 -1 -1
-4 -4 -4 -4

Asp

Cys

Gln

Glu

Gly

His

Ile

Leu

Lys

Met

Phe

Pro

Ser

A
A

C
C

G
G

I
I

L
L

M
M

P
P

S
H

V
V

Y
Y

Val

Tyr

Trp

Substitutions between acidic residues

Thr

Hydrophobic
Aromatic
Polar
Basic
Acidic

4
-1
-2
-2
0
-1
-1
0
-2
-1
-1
-1
-1
-2
-1
1
0
-3
-2
0
-2
-1
0
-4

Asn

A
R
N
D
C
Q
E
G
H
I
L
K
M
F
P
S
T
W
Y
V
B
Z
X
*

Arg

Ala
Arg
Asn
Asp
Cys
Gln
Glu
Gly
His
Ile
Leu
Lys
Met
Phe
Pro
Ser
Thr
Trp
Tyr
Val

Ala

BLOSUM62 matrix

5
0
-2
-3
1
0
-2
0
-3
-2
2
-1
-3
-2
-1
-1
-3
-2
-3
-1
0
-1
-4

6
1
-3
0
0
0
1
-3
-3
0
-2
-3
-2
1
0
-4
-2
-3
3
0
-1
-4

6
-3
0
2
-1
-1
-3
-4
-1
-3
-3
-1
0
-1
-4
-3
-3
4
1
-1
-4

9
-3
-4
-3
-3
-1
-1
-3
-1
-2
-3
-1
-1
-2
-2
-1
-3
-3
-2
-4

5
2
-2
0
-3
-2
1
0
-3
-1
0
-1
-2
-1
-2
0
3
-1
-4

5
-2
0
-3
-3
1
-2
-3
-1
0
-1
-3
-2
-2
1
4
-1
-4

6
-2
-4
-4
-2
-3
-3
-2
0
-2
-2
-3
-3
-1
-2
-1
-4

8
-3
-3
-1
-2
-1
-2
-1
-2
-2
2
-3
0
0
-1
-4

4
2
-3
1
0
-3
-2
-1
-3
-1
3
-3
-3
-1
-4

4
-2
2
0
-3
-2
-1
-2
-1
1
-4
-3
-1
-4

5
-1
-3
-1
0
-1
-3
-2
-2
0
1
-1
-4

5
0
-2
-1
-1
-1
-1
1
-3
-1
-1
-4

6
-4
-2
-2
1
3
-1
-3
-3
-1
-4

7
-1
-1
-4
-3
-2
-2
-1
-2
-4

4
1
-3
-2
-2
0
0
0
-4

5
-2
-2
0
-1
-1
0
-4

11
2
-3
-4
-3
-2
-4

7
-1
-3
-2
-1
-4

4
-3 4
-2 1 4
-1 -1 -1 -1
-4 -4 -4 -4

Asp

Cys

Gln

Glu

Gly

His

Ile

Leu

Lys

Met

Phe

Pro

Ser

A
A

C
C

G
G

I
I

L
L

M
M

P
P

Val

S
H

V
V

Y
Y

Tyr

Trp

Substitutions between basic residues

Thr

Hydrophobic
Aromatic
Polar
Basic
Acidic

4
-1
-2
-2
0
-1
-1
0
-2
-1
-1
-1
-1
-2
-1
1
0
-3
-2
0
-2
-1
0
-4

Asn

A
R
N
D
C
Q
E
G
H
I
L
K
M
F
P
S
T
W
Y
V
B
Z
X
*

Arg

Ala
Arg
Asn
Asp
Cys
Gln
Glu
Gly
His
Ile
Leu
Lys
Met
Phe
Pro
Ser
Thr
Trp
Tyr
Val

Ala

BLOSUM62 matrix

5
0
-2
-3
1
0
-2
0
-3
-2
2
-1
-3
-2
-1
-1
-3
-2
-3
-1
0
-1
-4

6
1
-3
0
0
0
1
-3
-3
0
-2
-3
-2
1
0
-4
-2
-3
3
0
-1
-4

6
-3
0
2
-1
-1
-3
-4
-1
-3
-3
-1
0
-1
-4
-3
-3
4
1
-1
-4

9
-3
-4
-3
-3
-1
-1
-3
-1
-2
-3
-1
-1
-2
-2
-1
-3
-3
-2
-4

5
2
-2
0
-3
-2
1
0
-3
-1
0
-1
-2
-1
-2
0
3
-1
-4

5
-2
0
-3
-3
1
-2
-3
-1
0
-1
-3
-2
-2
1
4
-1
-4

6
-2
-4
-4
-2
-3
-3
-2
0
-2
-2
-3
-3
-1
-2
-1
-4

8
-3
-3
-1
-2
-1
-2
-1
-2
-2
2
-3
0
0
-1
-4

4
2
-3
1
0
-3
-2
-1
-3
-1
3
-3
-3
-1
-4

4
-2
2
0
-3
-2
-1
-2
-1
1
-4
-3
-1
-4

5
-1
-3
-1
0
-1
-3
-2
-2
0
1
-1
-4

5
0
-2
-1
-1
-1
-1
1
-3
-1
-1
-4

6
-4
-2
-2
1
3
-1
-3
-3
-1
-4

7
-1
-1
-4
-3
-2
-2
-1
-2
-4

4
1
-3
-2
-2
0
0
0
-4

5
-2
-2
0
-1
-1
0
-4

11
2
-3
-4
-3
-2
-4

7
-1
-3
-2
-1
-4

4
-3 4
-2 1 4
-1 -1 -1 -1
-4 -4 -4 -4

Asp

Cys

Gln

Glu

Gly

His

Ile

Leu

Lys

Met

Phe

Pro

Ser

A
A

C
C

G
G

I
I

L
L

M
M

P
P

Val

S
H

V
V

Y
Y

Tyr

Trp

Substitutions between aromatic residues

Thr

Hydrophobic
Aromatic
Polar
Basic
Acidic

4
-1
-2
-2
0
-1
-1
0
-2
-1
-1
-1
-1
-2
-1
1
0
-3
-2
0
-2
-1
0
-4

Asn

A
R
N
D
C
Q
E
G
H
I
L
K
M
F
P
S
T
W
Y
V
B
Z
X
*

Arg

Ala
Arg
Asn
Asp
Cys
Gln
Glu
Gly
His
Ile
Leu
Lys
Met
Phe
Pro
Ser
Thr
Trp
Tyr
Val

Ala

BLOSUM62 matrix

5
0
-2
-3
1
0
-2
0
-3
-2
2
-1
-3
-2
-1
-1
-3
-2
-3
-1
0
-1
-4

6
1
-3
0
0
0
1
-3
-3
0
-2
-3
-2
1
0
-4
-2
-3
3
0
-1
-4

6
-3
0
2
-1
-1
-3
-4
-1
-3
-3
-1
0
-1
-4
-3
-3
4
1
-1
-4

9
-3
-4
-3
-3
-1
-1
-3
-1
-2
-3
-1
-1
-2
-2
-1
-3
-3
-2
-4

5
2
-2
0
-3
-2
1
0
-3
-1
0
-1
-2
-1
-2
0
3
-1
-4

5
-2
0
-3
-3
1
-2
-3
-1
0
-1
-3
-2
-2
1
4
-1
-4

6
-2
-4
-4
-2
-3
-3
-2
0
-2
-2
-3
-3
-1
-2
-1
-4

8
-3
-3
-1
-2
-1
-2
-1
-2
-2
2
-3
0
0
-1
-4

4
2
-3
1
0
-3
-2
-1
-3
-1
3
-3
-3
-1
-4

4
-2
2
0
-3
-2
-1
-2
-1
1
-4
-3
-1
-4

5
-1
-3
-1
0
-1
-3
-2
-2
0
1
-1
-4

5
0
-2
-1
-1
-1
-1
1
-3
-1
-1
-4

6
-4
-2
-2
1
3
-1
-3
-3
-1
-4

7
-1
-1
-4
-3
-2
-2
-1
-2
-4

4
1
-3
-2
-2
0
0
0
-4

5
-2
-2
0
-1
-1
0
-4

11
2
-3
-4
-3
-2
-4

7
-1
-3
-2
-1
-4

4
-3 4
-2 1 4
-1 -1 -1 -1
-4 -4 -4 -4

Asp

Cys

Gln

Glu

Gly

His

Ile

Leu

Lys

Met

Phe

Pro

Ser

A
A

C
C

G
G

I
I

L
L

M
M

P
P

Val

S
H

V
V

Y
Y

Tyr

Trp

Substitutions
between polar
residues

Thr

Hydrophobic
Aromatic
Polar
Basic
Acidic

4
-1
-2
-2
0
-1
-1
0
-2
-1
-1
-1
-1
-2
-1
1
0
-3
-2
0
-2
-1
0
-4

Asn

A
R
N
D
C
Q
E
G
H
I
L
K
M
F
P
S
T
W
Y
V
B
Z
X
*

Arg

Ala
Arg
Asn
Asp
Cys
Gln
Glu
Gly
His
Ile
Leu
Lys
Met
Phe
Pro
Ser
Thr
Trp
Tyr
Val

Ala

BLOSUM62 matrix

5
0
-2
-3
1
0
-2
0
-3
-2
2
-1
-3
-2
-1
-1
-3
-2
-3
-1
0
-1
-4

6
1
-3
0
0
0
1
-3
-3
0
-2
-3
-2
1
0
-4
-2
-3
3
0
-1
-4

6
-3
0
2
-1
-1
-3
-4
-1
-3
-3
-1
0
-1
-4
-3
-3
4
1
-1
-4

9
-3
-4
-3
-3
-1
-1
-3
-1
-2
-3
-1
-1
-2
-2
-1
-3
-3
-2
-4

5
2
-2
0
-3
-2
1
0
-3
-1
0
-1
-2
-1
-2
0
3
-1
-4

5
-2
0
-3
-3
1
-2
-3
-1
0
-1
-3
-2
-2
1
4
-1
-4

6
-2
-4
-4
-2
-3
-3
-2
0
-2
-2
-3
-3
-1
-2
-1
-4

8
-3
-3
-1
-2
-1
-2
-1
-2
-2
2
-3
0
0
-1
-4

4
2
-3
1
0
-3
-2
-1
-3
-1
3
-3
-3
-1
-4

4
-2
2
0
-3
-2
-1
-2
-1
1
-4
-3
-1
-4

5
-1
-3
-1
0
-1
-3
-2
-2
0
1
-1
-4

5
0
-2
-1
-1
-1
-1
1
-3
-1
-1
-4

6
-4
-2
-2
1
3
-1
-3
-3
-1
-4

7
-1
-1
-4
-3
-2
-2
-1
-2
-4

4
1
-3
-2
-2
0
0
0
-4

5
-2
-2
0
-1
-1
0
-4

11
2
-3
-4
-3
-2
-4

7
-1
-3
-2
-1
-4

4
-3 4
-2 1 4
-1 -1 -1 -1
-4 -4 -4 -4

Asp

Cys

Gln

Glu

Gly

His

Ile

Leu

Lys

Met

Phe

Pro

Ser

A
A

C
C

G
G

I
I

L
L

M
M

P
P

S
H

V
V

Y
Y

Val

Tyr

Trp

Substitutions between hydrophobic residues

Thr

Hydrophobic
Aromatic
Polar
Basic
Acidic

4
-1
-2
-2
0
-1
-1
0
-2
-1
-1
-1
-1
-2
-1
1
0
-3
-2
0
-2
-1
0
-4

Asn

A
R
N
D
C
Q
E
G
H
I
L
K
M
F
P
S
T
W
Y
V
B
Z
X
*

Arg

Ala
Arg
Asn
Asp
Cys
Gln
Glu
Gly
His
Ile
Leu
Lys
Met
Phe
Pro
Ser
Thr
Trp
Tyr
Val

Ala

BLOSUM62 matrix

RBLOSUM62
BLOSUM62 miscalculations improve search performance
MP Styczynski, KL Jensen, I Rigoutsos, G Stephanopoulos
Nat. Biotech. 26: 274275, 2008

The BLOSUM family of substitution matrices, and particularly BLOSUM62, is the de

facto standard in protein database searches and sequence alignments. In the course
of analyzing the evolution of the Blocks database, we noticed errors in the software
source code used to create the initial BLOSUM family of matrices (available online at
ftp://ftp.ncbi.nih.gov/repository/blocks/ unix/blosum/blosum.tar.Z). The result of these
errors is that the BLOSUM matricesBLOSUM62, BLOSUM50, etc.are quite
different from the matrices that should have been calculated using the algorithm
described by Henikoff and Henikoff. Obviously, minor errors in research, and
particularly in software source code, are quite common. This case is noteworthy for
three reasons: first, the BLOSUM matrices are ubiquitous in computational biology;
second, these errors have gone unnoticed for 15 years; and third, the incorrect
matrices perform better than the intended matrices.

RBLOSUM62
BLOSUM62 miscalculations improve search performance
MP Styczynski, KL Jensen, I Rigoutsos, G Stephanopoulos
Nat. Biotech. 26: 274275, 2008
Supplementary Figure 5. The revised
BLOSUM matrix, RBLOSUM62.
Values in red are one greater than they
would be in BLOSUM62, while values in
green are one less than they would be in
BLOSUM62.
The entropy of this matrix (based on raw
matrix values) is 0.6626 bits.

PAM vs BLOSUM matrices

BLOSUM62 scores - PAM160 scores

BLOSUM62 score matrix

Both matrices have identical relative entropies (0.70)

Source: Henikoff & Henikoff, 1992

PAM vs BLOSUM matrices

A comparison of the matrices can be done on the basis of their
"information content" (see precise definition later)

more conserved
proteins

PAM100 BLOSUM90
PAM120 BLOSUM80
PAM160 BLOSUM60

more distant
proteins

PAM200 BLOSUM52
PAM250 BLOSUM45

PAM/BLOSUM matrices: link with real time?

It is very difficult to relate the substitution matrices with the real
evolution time because the rate of mutations depends
- on the time
- on the species
- on the protein type
Nevertheless, using 57 enzymes from various
organisms (animals, plants, fungi, bacteria) and
under simplifying assumptions, Doolittle et al
(1996) could relate the scores obtained with
PAM250/BLOSUM62 matrices (converted into a
distance) with the evolution time.
Using a linear fitting, they estimated that
eukaryotes and eubacteria last shared a
common ancestor about 2 billion (2.109) years
ago.

Doolittle et al (1996) Determining divergence time of the major kingdoms of

living organisms with a protein clock. Science 271: 470-477.

GONNET matrix
GONNET scoring matrix

Gonnet, Cohen, Benner (1992). Exhaustive

matching of the entire protein sequence database.
Science. 256:1443-1445.

A different method to measure differences

among amino acids was developed by
Gonnet, Cohen and Benner (1992) using
exhaustive pairwise alignments of the
protein databases as they existed at that
time. They used classical distance
measures to estimate an alignment of the
proteins. They then used this data to
estimate a new distance matrix. This was
used to refine the alignment, estimate a
new distance matrix and so on iteratively.
They noted that the distance matrices (all
first normalised to 250 PAMs) differed
depending on whether they were derived
from distantly or closely homologous
proteins. They suggest that for initial
comparisons their resulting matrix should
be used in preference to a PAM250 matrix,
and that subsequent refinements should
be done using a PAM matrix appropriate to
the distance between proteins.

Source: http://www.ebi.ac.uk/help/matrix.html

GONNET matrix
GONNET scoring matrix

Protein
Database

pair-wise
alignments

new matrix

PAM250

iterative
pair-wise
alignments

GONNET m atrix

Other substitution matrices for amino acids

Many other families of substitution matrices for amino acids have
been proposed:
Simple identity matrices
Matrices based on the genetic code changes (score the minimum of
nucleotide changes to change a codon for one amino acid into a codon
for another) (Fitch, 1966)
Matrices based on chemical similarities of amino acid side chains
(molecular volume, polarity, hydrophobicity) (Vogt et al, 1995)
Matrices based on structurally aligned 3D structures (Risler et al, 1998;
Henikoff & Henikoff, 1993)
Dipeptide substitution matrices (Gonnet et al, 1994)
Specific substitution matrices for transmembrane proteins (Jones et al,
1994)
Source: Mount, 2004

Substitution matrices for nucleotides

Substitution matrices for nucleotides have also
been obtained in a similar way than the PAM
matrices for proteins

purines

pyrimidines

C
transition

transversion

PAM10

90.7

-3.70

-2.19

-3.70

-3.70

90.7

-3.70

-2.19

-2.19

-3.70

90.7

-3.70

-3.70

-2.19

-3.70

90.7

States et al (1991) Improved sensitivity of nucleic acid database searches

using application-specific scoring matrices. Methods 3: 66-70.

Substitution matrices for nucleotides

Probability matrix for nucleotides
A

0.99

0.00333 0.99

0.00333 0.00333 0.99

0.00333 0.00333 0.00333 0.99

0.99

0.006

0.99

0.002

0.002

0.99

0.002

0.002

0.006

Probability matrix based on a model of

uniform mutation rates among
nucleotides

Probability matrix based on a model of

3-fold higher transition (substitution
between the purines A and G or pyrimidine
C and T) that transversion (substitution
form a purine to a pyrimidine or a from
pyrimidine to a purine) (Li & Graur, 1991)

0.99
Source: Mount, 2004

Substitution matrices for nucleotides

Probability matrix for nucleotides
A

0.99

0.00333 0.99

0.00333 0.00333 0.99

0.00333 0.00333 0.00333 0.99

Exercise
Here are given the probability matrices.

0.99

0.006

0.99

0.002

0.002

0.99

0.002

0.002

0.006

0.99

What would be the corresponding

scoring matrices (PAM1-like) if we
assume a equal probability of each
nucleotide (fi = 0.25)? Give the log-odd
score and use log2.
What would be the PAM-like scoring
matrices corresponding to a distance of
2 PAM?
What would be the PAM-like scoring
matrices corresponding to a distance of
10 PAM? of 100 PAM?

Substitution matrices for nucleotides

Scoring matrix for nucleotides
A

-6

-6

-6

-6

-6

-6

-5

-7

-7

-7

-7

-5

Scoring matrix based on a model of

uniform mutation rates among
nucleotides (PAM1-like matrix)

Scoring matrix based on a model of 3-

fold higher transition that transversion
(PAM1-like matrix)

2
Source: Mount, 2004

Substitution matrices for nucleotides

Scoring matrix for nucleotides
A

-5

-5

-5

-5

-5

-5

-4

-6

-6

-6

-6

-4

Scoring matrix based on a model of

uniform mutation rates among
nucleotides (PAM2-like matrix)

Scoring matrix based on a model of 3-

fold higher transition that transversion
(PAM2-like matrix)

2
Source: Mount, 2004

Substitution matrices for nucleotides

Scoring matrix for nucleotides
A

1.86

-3.01

1.86

-3.01

-3.01

1.86

-3.01

-3.01

-3.01

1.86

1.86

-2.18

1.86

-3.70

-3.70

1.86

-3.70

-3.70

-2.18

Scoring matrix based on a model of

uniform mutation rates among
nucleotides (PAM10-like matrix)

Scoring matrix based on a model of 3-

fold higher transition that transversion
(PAM10-like matrix)

1.86
Source: Mount, 2004

Substitution matrices for nucleotides

Scoring matrix for nucleotides
A

0.83

-0.46

0.83

-0.46

-0.46

0.83

-0.46

-0.46

-0.46

0.83

0.88

0.069

0.88

-0.86

-0.86

0.88

-0.86

-0.86

0.069

Scoring matrix based on a model of

uniform mutation rates among
nucleotides (PAM100-like matrix)

Scoring matrix based on a model of 3-

fold higher transition that transversion
(PAM100-like matrix)
Note the positive values obtained
for the transitions A-G and C-T!

0.88
Source: Mount, 2004

Substitution matrices for nucleotides

Scoring matrix for nucleotides
A

-6

-6

-6

-6

-6

-6

-5

-7

-7

-7

-7

-5

Here are scoring PAM-1 matrices. By

multiplying the probability matrix with
itself, one can easily get PAM-n
matrices.

Exercise

Calculate the correspondance between

those PAM-n matrices and the percent
of identity.

Substitution matrices for nucleotides

Correspondance between the PAM distance and the identity level
identity
%

difference
%

PAM index
(n)

99

90.6

9.4

10

78.7

21.3

25

63.5

36.5

50

44.8

55.2

100

identity
%

difference
%

99

90.7

9.3

10

79.0

21.0

25

64.2

35.8

50

46.3

53.7

100

PAM index
(n)

model of uniform mutation rates among

nucleotides
Note the mismatch scores in this model tend to 0 as
PAM distance increases.
Thus, this matrix is not very informative at high PAM
distance and should be used only for comparing
sequences that are quite similar (using a low index
PAM matrix).

model of 3-fold higher transition that

transversion
Note that as PAM distance increases, the mismatch
scores in this model become positive and appear as
conservative s ubstitutions!
Thus, this model can provide much more information
than the uniform mutation model and should be used
for distantly related sequences.
Source: Mount, 2004

Substitution matrices for nucleotides

Other substitution matrices for nucleotides have been derived...
Chiaromonte et al (2002) have obtained matrices by analysis of alignments
of distinct regions of the human genome with different G+C content.

CFTR region
(37% G+C)

HOXD region
(47% G+C)

hum16pter region
(53% G+C)

Source: Zvelebil & Baum, 2007

Substitution matrices for nucleotides

Nucleotide scoring matrices used in FASTA and BLAST
FASTA and BLAST uses arbitrary nucleotide matrices

FASTA and WU-BLAST

NCBI-BLAST

-4

-4

-4

-4

-4

-4

should be used to detect homologous

alignments that are 65% identical
(edge of the twilight zone).

-2

-2

-2

-2

-2

-2

should be used to detect homologous

alignments that are 95% identical
(almost perfect match)

Source: Eddy, 2004

Substitution matrices: summary

Substitution matrices allow to detect similarities between more
distant proteins than what would be detected with the simple
identity of residues.

Different substitution scoring matrices have been established

PAM (Dayhoff, 1979)
BLOSUM (Henikoff & Henikoff, 1992)
Residue categories (Phylip)
...

Limitations of the substitution scoring matrices

They assumed independance between successive residues
They have been derived from manually aligned sequences
They have been built from a limited data set

Substitution matrices: summary

The matrix must be chosen carefully, depending on the expected rate of
conservation between the sequences to be aligned.
Beware
With PAM matrices
The score indicates the percentage of substitution per position
=> higher index are appropriate for more distant proteins
With BLOSUM matrices
The score indicates the percentage of conservation
=> higher index are appropriate for more conserved proteins

Matrices used in BLAST

Source: NCBI

Substitution matrices: further reading

PAM
Dayhoff et al. (1978). A model of evolutionary change in proteins. In Atlas of Protein Sequence and
Structure, vol. 5, suppl. 3, 345352. National Biomedical Research Foundation, Silver Spring, MD,
1978.

BLOSUM
Henikoff & Henikoff (1992). Amino acid substitution matrices from protein blocks. PNAS 89:10915-
10919.

Others substitution matrices for amino acids

Fitch (1966) An improved method of testing for evolutionary homology, J Mol Biol 21: 112-125.
Vogt et al (1995) An assessment of amino acid exchange matrices: the twilight zone re-visited. J Mol
Biol 249: 816-831.
Risler et al (1988) Amino acid substitution in structurally related proteins: a pattern recognition
approach, J Mol Biol 204: 1019-1029.
Henikoff & Henikoff (1993).Performance evaluation of amino acid substitution matrices, Prot Struct
Funct Genet 17: 49-61.
Gonnet (1994) Analysis of amino acid substitution during divergent evolution: the 400 x 400 dipetide
substitution matrix, Biochem Biophys Res Commun 199: 489-496.
Jones (1994) A mutational data matrix for transmembrane proteins, FEBS Lett. 339: 269-275.

Substitution matrices: further reading

Substitution matrices for nucleotides
States et al (1991) Improved sensitivity of nucleic acid database searches using application-specific
scoring matrices. Methods 3: 66-70.
Chiaromonte et al (2002) Scoring pair-wise genomic sequence alignments. Pac Symp Biocomput 7:
115-126.

Comparison and discussion of substitution matrices

Doolittle (1981) Similar amino acid sequences: chance or common ancestry? Science 214: 149-159
Henikoff & Henikoff (1993). Performance evaluation of amino acid substitution matrices, Prot Struct
Funct Genet 17: 49-61.
Pearson (1995) Comparison of methods for searching proteins sequence databases. Protein Sci 4:
1150-1160.
Pearson (1996) Effective protein sequence comparison. Meth. Enzymol 266: 227-258.
Altschul (1991) Amino acid substitution matrices from an information theoretic perspective. J Mol
Biol. 219:555-65.

Reviews
Eddy (2004) Where did the BLOSUM62 alignment score matrix come from? Nature Biotech 22:
1035-1036.