QualScore--training set distribution unusable.

[pagi...@gmail.com]

Hi:

There are some problems when I using the QualScore.

Enumerating scans:
processing LT20060823_P0031_1_07.mzXML...
processing LT20060823_P0031_3_09.mzXML... 500
processing LT20060823_P0031_2_08.mzXML...
processing LT20060823_P0031_4_10.mzXML... 1000 1500 1504
Features calculated for: 1504 (40519 ms)
Training set: 530/0 (good/bad)
Unassigned spectra: 291
ERROR: Training set distribution unusable.
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
I don't know why?
Could you help me?
Thank you~

By the way, it is hard to find this software. @_@
and i still can't find "ISB/SPC Proteomic Tools -- TPP Web
Interface"!!

HanYin

[James Eddes]

Hi HanYin

QualScore needs both good and bad assignemtns to train the spectrum classifier. By default, PeptideProphet automatically excludes assignments with probability below 0.05.

You must re-run PeptideProphet on your dataset with the probability threshold set to 0. This can be done either through Petunia or using the -p0 argument when running xinteract from the command line.

Instructions for downloading the software can be found by searching for the spctools-discuss forum for "qualscore download" or from this post:
http://groups.google.com/group/spctools-discuss/browse_thread/thread/c8faa616739b320/b8888373789767fb?lnk=gst&q=Qualscore+download&rnum=5#b8888373789767fb <http://groups.google.com/group/spctools-discuss/browse_thread/thread/c8faa616739b320/b8888373789767fb?lnk=gst&q=Qualscore+download&rnum=5#b8888373789767fb>

Please let me know if you are still having difficulty running the tool.

Good luck,
James

________________________________

[Momo]

Hi, James:

I have set -p0 when I use xinteract, but I still got these message

======================================================
XML
Loading interact data:
831
Interact data file contains 830 entries.
92 +1 were removed.
Enumerating scans:
processing jw_16c.mzXML...
processing jw_19cr.mzXML...
processing jw_17cr.mzXML...
processing jw_22c.mzXML...
processing jw_21cr.mzXML...
processing jw_21c.mzXML...
processing jw_18c.mzXML...
processing jw_17c.mzXML... 500
processing jw_20c.mzXML...
processing jw_18cr.mzXML... 738
Features calculated for: 738 (22893 ms)
Training set: 597/5 (good/bad)
Unassigned spectra: 24

ERROR: Training set distribution unusable.

=============================================================

But if I use different input file, sometimes it can run successful
Does it mean that these spectra in my files are too less to model the
distribution?
Could you help me ?
Thank you :)

Momo

On Jul 12, 4:10 pm, "James Eddes" <jed...@systemsbiology.org> wrote:
> Hi HanYin
>
> QualScore needs both good and bad assignemtns to train the spectrum classifier. By default, PeptideProphet automatically excludes assignments with probability below 0.05.
>
> You must re-run PeptideProphet on your dataset with the probability threshold set to 0. This can be done either through Petunia or using the -p0 argument when running xinteract from the command line.
>

> Instructions for downloading the software can be found by searching for the spctools-discuss forum for "qualscore download" or from this post:http://groups.google.com/group/spctools-discuss/browse_thread/thread/...<http://groups.google.com/group/spctools-discuss/browse_thread/thread/...>

>
> Please let me know if you are still having difficulty running the tool.
>
> Good luck,
> James
>
> ________________________________
>

> From: spctools...@googlegroups.com on behalf of pagi0...@gmail.com
> Sent: Thu 7/12/2007 3:41 PM
> To: spctools-discuss
> Subject: QualScore--training set distribution unusable.
>
> Hi:
>
> There are some problems when I using the QualScore.
>
> Enumerating scans:
> processing LT20060823_P0031_1_07.mzXML...
> processing LT20060823_P0031_3_09.mzXML... 500
> processing LT20060823_P0031_2_08.mzXML...
> processing LT20060823_P0031_4_10.mzXML... 1000 1500 1504
> Features calculated for: 1504 (40519 ms)
> Training set: 530/0 (good/bad)
> Unassigned spectra: 291
> ERROR: Training set distribution unusable.
> ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
> I don't know why?
> Could you help me?
> Thank you~
>
> By the way, it is hard to find this software. @_@
> and i still can't find "ISB/SPC Proteomic Tools -- TPP Web
> Interface"!!
>
> HanYin
>

> winmail.dat
> 7KDownload

[James Eddes]

Hi Momo

Yes, it means that there are not enough low probability assignments to
form a training set. The -p0 option should have included all of the low
probability hits, which it appears to have done. However, I would have
expected there to be many more low probability assignments from a
dataset comprised of 10 mzXML files.

What really puzzles me is that I wouldn't expect PeptideProphet to
properly derive probabilities from such a mix. I recommend that you
inspect the PeptideProphet probability models for anything unusual.
Take note of the number of spectra listed on the peptideprophet summary
page.

- James