[Bioperl-l] 2 psi-blast questions

[Donald G. Jackson]

Hadas,

the answer is 'yes and no'.

You can set the -h value ($blaster->h()) to adjust the inclusion 
threshold for refining the scoring matrix.  However, it only takes 
expect values - not identity/length - and the expect value can't go 
below zero.  Lower e values are more significant; e=0 roughly means 
there isn't a snowball's chance in the netherworld of getting that hit 
by chance in a dataset the size of the one you searched. 

You can use the Bio::SearchIO::Blast parser's hit_filter and hsp_filter 
methods to pass in a coderef that rejects hits above/below a certain 
identity and length.
If you do so, make sure you call identity on the HSP result objects, NOT 
the Hit objects as the identity numbers from Hit are inaccurate (a BLAST 
issue, not a SearchIO issue). 

I find that running PsiBlast takes some tweaking - adjusting the values 
for -e, -h, and -j (# rounds) so I build a good matrix and the search 
doesn't converge.  Also look at -b and -v  (# alignments/descriptions 
returned) to make sure you go deep enough into the results.

Hope this helps,

Don Jackson

Hadas Leonov wrote:

> Hi,
> I noticed that a psi-blast report only considers new hits if the 0.0 < 
> E-value < 0.01 (?).
> Is there any possibility to change these parameters ? especially the 
> 0.0 limit, cause large query sequences with 60% identity might still 
> give an e value of 0.0, thereby causing me to miss them completely by 
> the standard (and comfortable) method of going through a psi blast 
> report:
>
> ...
> $result = $psi_report->round($iter);
> $newHits_ref = $result->newhits;
> HIT: while($hit = $result->nextSbjct) {
> my $hitName = $hit->name;
> $is_new = grep /\Q$hitName\E/, @{$newHits_ref};
> unless ($is_new ) {next HIT;}
> #do something with the new hit
>
> }
> ...
>
> furthermore, the line : $hit = $result->nextSbjct
> sometimes causes the following error message to be displayed -
> -------------------- WARNING ---------------------
> MSG: Possible error (2) while parsing BLAST report!
> ---------------------------------------------------
> but i don't seem to find anything wrong with the actual parsing result.
>
> any ideas?
>
> Thanks,
> Hadas.
>
>------------------------------------------------------------------------
>
>_______________________________________________
>Bioperl-l mailing list
>Bioperl-l at portal.open-bio.org
>http://portal.open-bio.org/mailman/listinfo/bioperl-l
>  
>