Transcript March9

Point Specific Alignment Methods
PSI – BLAST
&
PHI – BLAST
In order to control the quality of the sequence matches in a BLAST search controls
are placed on the E – value of the result.
The Expect value (E) is a parameter that describes the number of hits one can "expect"
to see just by chance when searching a database of a particular size. It decreases
exponentially with the Score (S) that is assigned to a match between two sequences.
Essentially, the E value describes the random background noise that exists for matches
between sequences. For example, an E value of 1 assigned to a hit can be interpreted
as meaning that in a database of the current size one might expect to see 1 match with
a similar score simply by chance. This means that the lower the E-value, or the closer
it is to "0" the more "significant" the match is. However, keep in mind that searches
with short sequences, can be virtually indentical and have relatively high E-value. This
is because the calculation of the E-value also takes into account the length of the
Query sequence. This is because shorter sequences have a high probability of
occurring in the database purely by chance.
One criticism of this type of control is that sequences having basically the same
functionality may be missed in the search since they score over the threshold
value. Here is one possible cure:
The Expect value can also be used as a convenient way to create a significance
threshold for reporting results. You can change the Expect value threshold on most
main BLAST search pages. When the Expect value is increased from the default value
of 10, a larger list with more low-scoring hits can be reported.
Another strategy is to change the reward/penalty ratio in the scoring system.
Many nucleotide searches use a simple scoring system that consists of a "reward" for
a match and a "penalty" for a mismatch. The (absolute) reward/penalty ratio should
be increased as one looks at more divergent sequences. A ratio of 0.33 (1/-3) is
appropriate for sequences that are about 99% conserved; a ratio of 0.5 (1/-2) is best
for sequences that are 95% conserved; a ratio of about one (1/-1) is best for
sequences that are 75% conserved.
On the other hand, if we become too liberal in expanding these parameters, or change
ratios without reason, we find that we can find matches for almost any sequence. For
example, consider the amino acid sequence (V was used in place of U):
CVTTHESTEAKWITHASHARPKNIFEELSESTRINGYMEATWILLRESULT
We will use the protien – protein BLAST for short sequences using a non-redundant
database, an Expect Value of 20 and the PAM30 matrix and the Smith-Wateman
algorithm.
We get 37 matches for this nonsense sequence. The highest scoring match has an E-value
of 1.3
gi|121705510|ref|XP_001271018.1| C6 transcription factor, put... 34.6 1.3
gi|58583535|ref|YP_202551.1| HmsF [Xanthomonas oryzae pv. ory... 33.7 2.3
gi|84625349|ref|YP_452721.1| HmsF protein [Xanthomonas oryzae... 33.7 2.3
gi|123445879|ref|XP_001311695.1| hypothetical protein TVAG_49... 32.9 4.2
gi|123469845|ref|XP_001318132.1| helicase, putative [Trichomo... 32.5 5.6
gi|118032193|ref|ZP_01503644.1| conserved hypothetical protei... 32.5 5.6
>gi|121705510|ref|XP_001271018.1|
[Aspergillus clavatus NRRL 1]
C6 transcription factor, putative
gi|119399164|gb|EAW09592.1|
C6 transcription factor, putative [Aspergillus
clavatus NRRL 1] Length=887 Score = 34.6 bits (74),
Expect = 1.3 Identities = 16/32 (50%), Positives = 19/32 (59%), Gaps = 9/32
(28%)
Query 26
EE---LSESTRINGYM----EATWI--LLRES 48
EE
L+ES+R
GYM
E TW+
L RES
Sbjct 223 EEDLNLTESSRATGYMGKNSELTWMQRLQRES 254
>gi|58583535|ref|YP_202551.1|
KACC10331]
HmsF [Xanthomonas oryzae pv. oryzae
gi|58428129|gb|AAW77166.1|
HmsF protein [Xanthomonas oryzae pv. oryzae
KACC10331] Length=663 Score = 33.7 bits (72),
Expect = 2.3 Identities = 16/41 (39%), Positives = 22/41 (53%), Gaps = 14/41
(34%)
Query 18 HARP--KNIFEELSESTRINGYMEATWIL------LRESVL 50
AR
K+I+E+L+
IN YME
IL
LR++ L
Sbjct 460 QARQIIKDIYEDLA----INSYMEG--ILFHDDGYLRDTEL 494
Even using Local Sequence Alignment Techniques and
Scoring Matrices such as high powers of PAM or low
values of BLOSUMn Database Searching may not find
what we want.
• Many homologous sequences share only limited sequence
identity.
• While they may adopt the same three-dimensional structure,
they may not have apparent similarity in pair wise alignments.
• Cases are known where BLAST and FASTA miss 10 – 20% of
“meaningful” hits.
• Scoring matrices do not accurately portray the similarity that
may exist within a particular family of proteins. They are tied to
a more general database.
In an attempt to correct this the idea of a Position Specific
Scoring Matrix (PSSM) was developed.
In PSI-BLAST the query sequence is subjected to a normal
BLAST search. From this a multiple-sequence alignment is
made between the query and all “significant” hits.
A new scoring matrix of size L rows and 20 columns is
derived using the frequency of the proteins within each
position of the alignment. (L is the length of the query
sequence.)
The previous example was taken from
Pevsner, J., Bioinformatics and Functional Genomics,
Wiley-LISS, 2003, p139
And involves a search with Query sequence RBP4 (NP_006735)
Here is a portion of the PSSM generated by Pevsner’s Search
Note Lines 6, 11, 12, 14, 15, 16, and 42 all of which are
scores for A against the 20 proteins.
The PSSM is then used as the query (not your original
sequence) to the database and another search to the
database.
The statistical significance of each match is estimated and
results are reported.
These last three steps are repeated iteratively until no new
sequences are reported that fall above the given significance
level or the user chooses to terminate the search.
A Schemematic of the PSI-Blast Process
Note the original query is not included in loop 2.
Pevsner reported the following data concerning his 2002
search with original query NP_006735
At this point we will do an update of these results by
going to http://www.ncbi.nlm.nih.gov/blast and choosing
the PSI-BLAST option with the default parameters.
A Dramatic Illustration of the Increased Sensitivity ot
PSI-BLAST Searching
PHI-BLAST stands for Pattern-Hit Initiated BLAST
Often it is the case that a protein of interest contains a signature
pattern of amino acids and residues that help to define it as part of the
family. This “signature” may be rather short in terms of its length
within the sequence, but it is important in defining a structural of
functional domain. It may even be the characteristic of an unknown
function as is the case in the following example:
Care must be taken to choose a pattern that is not common
within the database. The algorithm only allows patterns that are
expected to occur at most once in every 5000 residues.
In the previous example the pattern is GXW where the X may
be any amino acid. Then we specify candidates for the
following amino acids [YF], [EA], or [IVLM]. These choices
are based on our observation of the test sequences and our
knowledge of the behavior of proteins (common protein
substitutions, hydrophobicity, etc.)
The database search is then performed looking for sequences
that contain the prescribed pattern.
Further iterations may be done based on this output using PSIBLAST which no longer uses the PHI pattern, but the PSSM
from the first report.
The output from the PHI-BLAST program is the same
as that of the PSI-BLAST program except that the
position of the pattern is highlighted in each of the
alignments .
The following alignment was obtained from an investigation of
immunoglobulin C-Region Domains:
We will investigate the conserved sequence: LXCLV using
PHI-BLAST.
Our starting point is with the Ig 2A C region of the mouse,
SwissProt Accession #P01865
We enter this information into the PHI-BLAST page
The first iteration of this search yields 31 new statistically significant hits.
One of these is given below. Note the *’s over the location of the pattern
LXCLV
Subsequent iterations are performed by PSI-BLAST independent of the
pattern. This search converged after 13 iterations.