PSI-BLAST2.132003/02/072008/09/10 15:43:47.529 GMT-5SusanCatesmscates@bioc.rice.eduSusanCatesmscates@bioc.rice.edubioinformaticsprotein database searchprotein motif profilePSI-BLASTsequence alignmentThis module is designed to introduce the student to PSI-BLAST. PSI-BLAST is a bioinformatics tool, offered through NCBI, for identifying weak, but biologically relevant sequence similarities.
PSI-BLAST (1)
(Position-Specific Iterated BLAST) is a tool that produces a position-specific scoring matrix constructed from a multiple alignment of the top-scoring BLAST responses to a given query sequence. This scoring matrix produces a profile designed to identify the key positions of conserved amino acids within a motif. When a profile is used to search a database, it can often detect subtle relationships between proteins that are distant structural or functional homologues. These relationships are often not detected by a BLAST search with a sample sequence query.
For an oversimplified example of what a consensus sequence, or profile, looks like, consider that the EF-hand binding loop of the calmodulin family could be represented as follows:
Loop Position # 1 3 4 5 6 8 12
Profile D x D G D/N G x I x x x E
Here "x" stands for positions where there is variability in amino acid type, and therefore, that position is not heavily weighted in the alignment. Comparing the profile to some actual binding loop sequences from different calmodulins is the best way to illustrate the derivation of this profile.
POSITION # 1 3 4 5 6 8 12
CALM_HUMAN_1 D K D G D G T I T T K E
CALF_NAEGR_1 D K D G D G T I T T S E
CALM_SCHPO_1 D R D Q D G N I T S N E
CALM_HUMAN_2 D A D G N G T I D F P E
CALF_NAEGR_2 D A D G N G T I D F T E
CALM_SCHPO_2 D A D G N G T I D F T E
CALM_HUMAN_3 D K D G N G Y I S A A E
CALF_NAEGR_3 D K D G N G F I S A Q E
CALM_SCHPO_3 D K D G N G Y I T V E E
CALM_HUMAN_4 D I D G D G Q V N Y E E
CALF_NAEGR_4 D I D G D N Q I N Y T E
CALM_SCHPO_4 D T D G D G V I N Y E E
The rules for deriving this simple profile are: 1) any position with 90% amino acid identity or greater is considered conserved in the profile, and thus a higher score would be given when the conserved amino acid is found at that position in the sequence, and 2) any position that always contains one of only two types of amino acids would be up-weighted to give a higher score whenever either of those two amino acids appears at that position. A program such as PSI-BLAST will employ more sophisticated rules to create a profile than this example, of course. It is easy to see even with these sequences that amino acid similarity could be taken into consideration in addition to amino acid identity, and exploited in the profile.
There are three common categories of homologues that are studied in relation to
biological molecules, sequence homology, structural homology, and functional
homology. Sequence homology is the easiest to identify, and is therefore the
primary target of many bioinformatics methods. Sequence homology yields direct
implications about the relatedness of proteins and their potential pathways of
derivation. However, to help understand how a protein is implicated in a
certain disease state, or how to design a pharmaceutical that interacts with a
given protein, functional and/or structural information is necessary.
Functional homologues are relatively easy to define, as they are any two
proteins, or protein domains, that perform similar functions. Structural
homologues contain similar "folds", which are localized regions of a molecule
that comprise a structural feature such as a "beta barrel" or "four helical
bundle" motif. The fold can encompass the entire protein, or just one domain
of the protein. A good introduction to the topic of protein folds can be found
at the website for the Internet Course on
The Principles of Protein Structure organized by
Birkbeck College (2).
When considering sequence, functional, or structural homology, it is important
to understand that one type of homology between proteins does not always infer
another type of homology. Nevertheless, it is a reasonable assumption that
proteins that are related through evolutionary pathways are likely to have
some degree of all three types of homology. PSI-BLAST was engineered to
identify distant relationships between sequences that are too subtle to
discover with a regular BLAST search.
In the first round, PSI-BLAST is just like a normal BLAST;
it finds sequence homologues. In the second round or "iteration" of PSI-BLAST,
it figures out which residues tend to be conserved by creating a custom profile
for each position of the sequence from a multiple alignment. Then another BLAST
is performed, using the profile to produce a position-specific scoring matrix
based on which positions evolution has conserved vs. which positions evolution
has allowed to vary. The sequences found after the first round are added to
the profile, allowing PSI-BLAST to detect more distant homologues in each
iteration.
One of the known weaknesses of PSI-BLAST is that its ability to detect distant
relationships between proteins is critically dependent on the choice of the
query sequence. For this reason, a recommended strategy with PSI-BLAST is to
query using individual functional domains. PSI-BLAST will then find other
proteins that share this domain, even if they do not possess overall homology.
To acquaint the new user with PSI-BLAST, this tutorial mimics an investigation
performed by
Aravind and Koonin (3) in 1999,
wherein new members of the HSP70/actin protein family were identified, except
the analysis in our tutorial will be on a much smaller scale than that
presented in the paper. The HSP70/actin family members were originally
recognized to have a common evolutionary origin as a result of a study
performed by
Bork et al. (4) in 1992.
A structural superposition of the structures of actin, hexokinase, and the
molecular chaperonin hsp70, and alignment of many sequences in each of the
three families, uncovered a set of common conserved residues, distributed in
five sequence motifs, that are involved in ATP binding and in a flexible
interdomain hinge. Although each of these proteins performs very different
functions, and their sequences are quite divergent, the similarity in the fold
of the ATP-binding domain is visually recognizable. These are all ATP-dependent
enzymes and the patterns discovered by Bork and associates could not be
detected by traditional BLAST-type sequence searches. Therefore, Aravind and
Koonin chose this family as a test of PSI-BLAST's ability to detect distant
evolutionary relationships.
Aravind and Koonin chose actin from the PDB file with accession code 1atn as
one of their query sequences. Begin the query by retrieving this sequence
from the PDB. Check the box for
searching the PDB archive that says "PDB ID", then enter the
accession code 1atn as the query. Notice that the crystal structure deposited
in this entry contained DNase I complexed with actin. There will be a link in
the menu in the blue border on the left entitled "FASTA Sequence", select
this link to download the sequence file. This file will contain
two sequences 1ATN:A (actin) and 1ATN:D (DNase I). Copy the sequence for
1ATN:A and paste it into the
BLAST query box that arises from choosing the Protein BLAST, then selecting "PSI-BLAST"
under the algorithm section. Change the database from "nr" to "swissprot",
but accept the default values for everything else. Click on BLAST, then view the results. NOTE THAT each time another iteration of PSI-BLAST runs, the results page will indicate the iteration number. This is very helpful for keeping track of the stage of the results.
There is a statistical section at the very end of the BLAST report. BLAST query sequence windows will be lined up with similar regions in the database, then the algorithm will try to extend the alignment with the database sequence in both directions along the query sequence. What is the number of successful extensions for the first BLAST report?
What is, by far, the most
common protein shown as a hit in the list of scores?
At the end of the list
of scores, note the section entitled "Sequences with E-values WORSE than threshold". Why is it that PSI-BLAST includes this section, but
BLAST does not?
Do any members from the Actin-like ATPase domain superfamily, other than actin
(or actin-like proteins), show up as hits in this section? If so, name one.
There are several buttons within the results document entitled "Run PSI-BLAST
iteration 2"; if they are difficult to locate, there is one at the end of the list
of descriptions and scores; click on one of these buttons
to run a second iteration of PSI-BLAST.
Look at the results window and above the graphical display,
it should say "Results of PSI-Blast iteration 2". Keep track of what the next iteration
number should be, and make sure it matches the iteration number that PSI-BLAST
displays to avoid getting lost within the PSI-Blast search. When the results appear,
look at the legend under the color alignment graph. It says that a yellow
starburst with the word "NEW" inside it indicates a new sequence that was
identified as a result of the most recent iteration, and a green dot indicates
a sequence that was already present prior to the most recent iteration.
What is the number of successful extensions for the second BLAST iteration?
Did the second iteration locate any new sequences
within (meaning less than or equal to) the threshold E-value? If so, how many?
Are there new members of the Actin-like ATPase domain
superfamily in the section entitled "Sequences with E-value WORSE than threshold"? If so, name one.
How many new sequences are identified as a result of the third iteration?
Are there new sequences returned that are within the
threshold E-value that are NOT actin or actin-like proteins? If so, name one.
Finally, perform the first iteration of the PSI-BLAST search with the same query
sequence as above, but with the database set to "nr".
What is the number of successful extensions for the first BLAST report when searching the nr database?
The "swissprot" database is a curated database, and all the descriptions on
the return list were informative. However, searching against the large "nr"
database yields a list of returns that should contain some proteins described
as "unknown", or "unnamed". The advantage of "nr" is that usually many more hits are
returned. Sometimes this is desirable, and sometimes it is not, but it is one thing
to consider when designing a PSI-BLAST search strategy.
A PSI-BLAST search can be forced to include more sequences in the profile at a
given iteration by altering the threshold. The default value is to accept any
results with E values less than 0.005. It is common to change this threshold
to 0.05 if the PSI-BLAST converges too quickly at 0.005. We did not take our
search to convergence, that would require continuing iterations until the last
iteration did not return any new hits. As this tutorial
illustrates, PSI-BLAST is a useful tool, but it often requires putting some
thought into the search strategy before it will produce meaningful results.
Altschul et al.Gapped Blast and PSI-Blast: a new generation of protein database search
programsNucleic Acids Research199725(17):3389-3402Internet Course on The Principles of
Protein Structure, a collaboration between Birkbeck College and the Virtual
School of Natural Sciences(VSNS) of the Globewide Network Academy (GNA) http://www.cryst.bbk.ac.uk/PPS95/ Aravind and Koonin Gleaning
non-trivial structural,functional and evolutionary information about proteins
by interative database searchesJMB 1999 287:1023-1040 Bork P, Sander C, Valencia A. An ATPase domain common to prokaryotic cell cycle proteins, sugar kinases,
actin, and hsp70 heat shock proteins Proc Natl Acad Sci U S A 1992 89(16):7290-4