[BiO BB] Re: PSI-BLAST query filter (Kevin Berney)

[Iddo Friedberg]

Hongyu Zhang wrote:

>It's true that PSI-BLAST perhaps was not designed to be used as a
>general purpose multiple alignment program as ClustalW was, but
>according to my observation lots of time its result is better than
>ClustalW. The problem of ClustalW is that, for a group of remote, and
>sometimes not so remote, homologous sequences, it often gives
>funny alignments. The situation can be worsened if the group has lots
>of sequences; meanwhile, PSI-BLAST alignment is much better in
>handling those situations. Again, for those who are not familiar with
>NCBI-BLAST, it provides the "-m 1" option to print out the multiple
>squence alignment instead of the default pair-wise alignment.
>  
>

The -m flag is cosmetic. The alignments you see are still  pairwise 
alignments to the query sequence, not a multiple alignment. -m 1  gives 
you the full sequences, as opposed to only the HSPs. (Note that HSPs may 
overlap, which can give you some interesting results when you use "-m 
1"). The reason PSI-BLAST "beavior" seems "better" is that it is 
performing pairwise alignments to a single sequence, so naturally the 
alignments would seem more ordered. ClustalW would give you "funny 
alignemnts" as beyond some point it is not equipped to handle long 
and/or low complexity and/or large numbers and/or distantly related  
sequences. T-COFFEE gives you more mileage in that respect, you might 
want to try that if you need a true MSA.

It all boils down to your biological question: whether it requires a 
true multiple alignment, for which the BLAST programs are not suited, or 
it just requires a bunch of sequences to "toe the line" of the original 
query.

>I assume most standard multiple sequence alignment methods are just
>multi-steps of pair-wise alignments in an optimized order, while
>profile methods are smarter in looking for some common
>patterns in an iterated fashion. I used a GUI version of
>ClustalW-based program a long long time ago (I forgot the
>program name), which also provides a profile option in its menu, but I
>didn't remember how well it works. Actually, some benchmarks on this
>topic would be nice.
>
>--
>Hongyu Zhang
>Computational biologist
>Ceres Inc.
>
>
>
>
>
>_______________________________________________
>Bioinformatics.Org general forum  -  BiO_Bulletin_Board at bioinformatics.org
>https://bioinformatics.org/mailman/listinfo/bio_bulletin_board
>
>
>  
>


-- 
Iddo Friedberg, Ph.D.
The Burnham Institute
10901 N. Torrey Pines Rd.
La Jolla, CA 92037 USA
Tel: +1 (858) 646 3100 x3516
http://ffas.ljcrf.edu/~iddo
==========================
The First Automated Protein Function Prediction SIG
Detroit, MI June 24, 2005
http://ffas.burnham.org/AFP