Scoring Matrices2.72003/02/272008/09/10 15:31:03.063 GMT-5SusanCatesmscates@bioc.rice.eduSusanCatesmscates@bioc.rice.edubioinformaticsBLOSUMPAMscoring matrixsequence alignmentsubstitution matrixThis module explains the general characteristics of the scoring matrices used to score sequence alignments. Guidelines are given to help the student choose which scoring scheme to choose for a given alignment, according to the characteristics of the sequences to be aligned.
In bioinformatics, scoring matrices for computing alignment scores are
often based on observed substitution rates, derived from the substitution
frequencies seen in multiple alignments of sequences. Every possible identity
and substitution is assigned a score based on the observed frequencies of such
occurences in alignments of related proteins. The score is calculated from the
frequency of occurrence of a match of the two individual amino acids in
evolutionarily related sequences, and provides a measure of a chance
alignment of the two amino acids. This score will also reflect the frequency
that a particular amino acid occurs in nature, as some amino acids are more
abundant than others. Higher scores indicate that the probability that those
two amino acids aligned by chance is very small, and lower scores indicate a
high probability the two amino acids aligned by chance, and are evolutionarily
unrelated. Thus, identities are assigned the most positive scores, frequently
observed substitutions also receive positive scores, but matches that are
unlikely to have been a result of evolution, and are more likely indicative
of unrelatedness at that position, are given negative scores. Matrices with
scoring schemes based on observed substitution rates are superior to simple
identity scores, or scores based solely on sidechain moiety similarity. The
two most commonly used types of scoring matrices are the
PAM matrices
and the BLOSUM matrices.
PAM (Percentage of Acceptable point Mutations per
108
years) matrices are based on
global alignments of closely related proteins.
The PAM 1 is the matrix calculated from comparisons of sequences with no more
than 1% divergence. Scores are derived from a mutation probability matrix
where each element gives the probability of the amino acid in column X
mutating to the amino acid in row Y after a particular evolutionary time,
for example after 1 PAM, or 1% divergence.
A PAM matrix is specific for a particular evolutionary distance, but may be
used to generate matrices for greater evolutionary distances by multiplying
it repeatedly by itself. However, at large evolutionary distances the
information present in the matrix is essentially degenerated. It is rare
that a PAM matrix would be used for an evolutionary distance any greater
than 256 PAMs.
Whereas the PAM matrices have been developed from global alignments,
the BLOSUM (BLOcks SUbstitution Matrix) matrices are based on local
multiple alignments of more distantly related sequences. For instance,
BLOSUM 62, the default matrix in BLAST, is a matrix calculated from
comparisons of sequences with no less than 62% identity. Unlike PAM matrices,
new BLOSUM matrices are never extrapolated from existing BLOSUM matrices,
but are always based on local multiple alignments. So, the BLOSUM 80 matrix
would be derived from a set of sequences having 80% sequence identity.
The level of relatedness of a set of sequences, therefore, directly effects
which scoring matrix is most appropriate for aligning the set, whether or
not it is a PAM or a BLOSUM matrix. Comparisons of closely related sequences
should use BLOSUM matrices with higher numbers and PAM matrices with lower
numbers. Conversely, BLOSUM matrices with low numbers and PAM matrices with high
numbers are preferable for comparisons of distantly related proteins. Nevertheless, a
single matrix may be reasonably efficient over a relatively
broad range of evolutionary change. The BLOSUM 62 matrix was chosen as the
default for BLAST as a result of an analysis by
Henikoff and Henikoff wherein BLOSUM 62
detected more distant
relationships in a BLAST search, and produced an alignment of diverged
proteins more in agreement with three-dimensional structures, than did
the corresponding PAM 60 matrix. The BLOSUM series does not include
any matrices suitable for very short query sequences, so, in these cases,
the PAM matrices may be used instead. Berkeley has a
Matrix Information website with a provisional table of
recommended substitution matrices and gap costs for shorter sequences.
Now, take a look at some scoring matrices. A
PAM Matrix website sponsored by Wageningen University, in the Netherlands, allows online computation of PAM matrices.
The default value is a PAM 250 matrix; calculate this matrix and look at
the results.
This PAM 250 matrix has a built-in gap penalty of -8, as seen in the * column.
There are 24 rows and 24 columns. Of course, the first 20 are the amino acids,
represented by the one letter code. B represents the case where there is
ambiguity between aspartate or asparigine, and Z is the case where there is
ambiguity between glutamate or glutamine. X represents an unknown, or
nonstandard amino acid.
In the PAM 250 matrix, where can the highest
scores for each amino acid be found? Why?
Would this be true for any
scoring matrix?
What row and column combination gives the highest score?
(Specify the score value.)
What is the second highest score?
(Specify the score value.)
Why are some scores for amino acid identities higher than others?
Use the back button on the browser, and calculate a PAM 100 matrix.
Are the two highest scoring matches the same combination of row
and column as in the PAM 250 matrix? (Discuss with a sentence or two.)
What is the gap penalty?
Explain any differences in the gap penalties of the
PAM 250 matrix versus the PAM 100 matrix.
To get an idea how the scoring matrix influences an alignment,
perform the following exercise using the
Biology Workbench. The Workbench will require a password (it's free), but it will grant entrance immediately upon registration of a password. Enter the site, and scroll down the page until the five menu buttons are visible. The "Session Tools" button allows the naming of a session, so that different jobs in progress can be saved under distinct sessions. Select "Session Tools", then select "Start New Session" and click on "Run" to change the name of "Default Session" to a new name. Once the workbench has been exited, the session will remain. Subsequently, clicking on the dot to the left of the session name under the "Session Tools" menu, and then selecting "Resume Session", will recall the session. The Workbench policy at the time of this writing is that old jobs are deleted only when an account has not been accessed for 6 months.
This tutorial will use sequences of hemoglobins (Hbs) from different
organisms to illustrate the properties of scoring matrices. Choose the
"Protein Tools" menu button, then choose the "Ndjinn Multiple Database Search"
from the menu at the bottom of the page. Biology Workbench has a large number of databases to search, for this exercise, click in the box to left of the database description to choose the "PDBFINDER" database.
Search the PDBFINDER database by typing in the PDB ID codes below into the search box at the top of the page. Import the sequences with the following PDB
ID codes (use the OR operator
between each PDB ID code to search for all of the records in the same search):
1T1N from trematomus newnesi (antarctic fish)1SPG from leiostomus xanthurus (spot croaker) 1QSI from homo sapiens (human)1IWH from equus cabullus (horse)1HV4 from anser indicus (goose)1HBR from gallus gallus (chicken)1H97 from paramphistomum epiclitum (trematode)1GVH from escherichia coli (enterobacteria)
The import function in the Workbench requires checking the boxes for all the PDB ID codes that were returned, then hitting the import button.
There will be several subunits returned with most of these sequences,
and some are duplicate sequences, so delete the following chains by
clicking the box on the left of the ID code and selecting "Delete Protein
Sequence(s)" from the pull-down menu at the bottom of the page:
1HV4_C1HV4_D1HV4_E1HV4_F1HV4_G1HV4_H1HBR_C1HBR_D1H97_B1QSI_C1QSI_D
After the above sequences have been deleted, choose "Select All Sequence(s)"
from the pull-down menu. Analyze the relatedness of this group of sequences
by selecting "ClustalW" from the pull-down menu to perform a multiple alignment
and draw a rooted cladogram. When the ClustalW page appears, before submitting
the alignment, scroll down the page and change the "Guide tree display:" to
"Rooted".
When the ClustalW results appear, first scroll down to the cladogram and
observe which of these sequences are most closely related versus the more
distant sequences.
Notice there are three separate clusters of branches descending from the root.
The two largest clusters are separated as a direct result of a structural
characteristic of hemoglobins.
What do each of these two clusters represent?
(If the answer is not immediately clear, read this
description of Hemoglobin from the University of Brescia's on-line
Biochemistry Course.)
According to this cladogram, what is the sequence that
is most closely related to human hemoglobin, ID code 1QSI?
According to this cladogram, what is the sequence that is most closely related to the E. coli flavoHb, ID code 1GVH?
According to this cladogram, what sequence is most closely related to the spot croaker
Hb, ID code 1SPG?
It is not as clear from the cladogram which sequences are
the most distantly related. However, scroll down past the cladogram to view
the ClustalW pairwise alignment scores.
Which two sequences yield the
lowest pairwise alignment score?
At the very bottom of the alignment page,
select "Import Alignments", to save this information for later reference,
should that be necessary. The imported alignments can only be viewed through
the "Alignment Tools" menu.
Apply the information elucidated by the multiple sequence alignment to test
the impact of varying the scoring matrices in pairwise alignments.
Return to "Protein Tools". Start with two sequences that are known to be
closely related, the human Hb chain B, 1QSI_B, and the horse Hb chain B,
1IWH_B, by checking the box to the left of each of their codes.
Choose "LALIGN" from the pull-down menu at the bottom of the page to compare
two protein sequences to each other with BLAST. When the LALIGN page appears,
next to select scoring matrix, choose PAM250 and run the alignment.
What is the (a) score of the alignment,
(b) the length of the alignment, and
(c) the percent identity?
Now, return to "Protein Tools" and run LALIGN again with the same two sequences,
1QSI_B and 1IWH_B, except choose the "PAM120" matrix
this time.
What is the (a) score of the alignment,
(b) the length of the alignment, and
(c) the percent identity?
(a) Which scoring matrix yielded the highest score for the alignment,
and why is this matrix the best choice for this alignment? (b) List any regions where the two alignments differ.
Return to "Protein Tools", this time selecting the 1HV4_A and the 1TIN_B
sequences, by checking the box next to their codes. Again, choose "LALIGN",
and perform an alignment with the default PAM250 matrix.
What is the (a) score of the alignment,
(b) the length of the alignment, and
(c) the percent identity?
Run LALIGN again on the same two sequences, using the
PAM120 matrix. What is the (a) score of the alignment,
(b) the length of the alignment, and
(c) the percent identity?
(a) Which scoring matrix yielded the highest score for the alignment,
and why is this matrix the best choice for this alignment? (b) List any regions where the two alignments differ.
Do the two different matrices always calculate
the same value for percent identity when the same 2 sequences are being
compared using each matrix?
Why or why not?
Most bioinformatics tools available on the web have selected default
scoring matrices that are based on a relatively exhaustive analysis
of which scoring schemes work best over a wide range of query sequence
characteristics.
However, it is important to not only know which scoring matrix is used
for a given alignment, but to consider the appropriateness of the
default matrix for a given query as well. It is a recurring theme of
bioinformatics that these computational tools
should not be treated as "black boxes" where one can ignore the internal
workings of the software,
but instead require thoughtful interaction on the part of the user.
Schwartz RM, Dayhoff MO Atlas of Protein Sequence and Structure, 5 suppl.Nat. Biomed. Res. Found.978 3:353-358 Washington D.C.Henikoff S, Henikoff JG. Amino acid substitution matrices from protein blocksProc Natl Acad Sci U S A. 1992 89(22):10915-9 Henikoff S, Henikoff JG. Performance evaluation of amino acid substitution matrices Proteins 1993 17(1):49-61