[ssml] psi-blast itteration...

[Mensur Dlakic]

Disclaimer: I am no BLAST expert, so some of the things I say below are 
just common(ly wrong) sense.

At 09:59 PM 8/30/2003 +0100, you wrote:
>Mensur Dlakic said:
> > Hi,
> >
> > I have seen the same thing happen many times with different versions 
> of  BLAST and
> > don't think it's a bug. Intuitively, I think this is because in  the 
> initial round
> > the sequence is scored using the substitution matrix  alone, while in 
> the second
> > iteration the profile is already built based on  that single sequence, and
> > naturally it comes closer to the sequence itself  than the substitution 
> matrix.
>
>
>Would the difference in e value be more prounounced in the second itteration
>if the first HSP had a low e value?

I think so. I routinely set E-value lower than h-value for PSI-BLAST 
iterations and often see that E-values improve quite a lot for many 
sequences, though not necessarily below the h-value so they don't get 
included in the profile.

>What additional information over the substitution matrix is used in the second
>itteration when we only use a single sequence in the profile?
>
>Does it work like this...
>
>1 take query, make profile
>2 query
>3 get hsps, rebuild profile
>4 query

I think that in the first iteration no profile is made until some hits are 
found, even if it's only a hit to itself.

>Is this a reason for including query sequences in the search space? (to track
>the value of self hits)?

I think that it has to be included, if for no other reason than only to 
have something for alignments. That said, I have seen that with some 
profiles the query sequence will stop being the best match after a few 
iterations, meaning that it'll move down the list a bit.

Best,

Mensur

==========================================================================
| Mensur Dlakic, Ph.D.              | Tel: (406) 994-2903                |
| Department of Microbiology        | Fax: (406) 994-4926                |
| Montana State University          | http://www.umich.edu/~mensur/      |
| 109 Lewis Hall, P.O. Box 173520   | http://myprofile.cos.com/mensur    |
| Bozeman, MT 59717-3520            | E-mail: mdlakic at montana.edu        |
==========================================================================