PeptideProphet - Kojak 1.5.5 vs. 1.6.1

[Thomas Gossenreiter]

Hi,

I have used Kojak 1.6.1 in the TPP pipeline and got some strange results (very similar ms2 spectra from one peptide with similar Kojak scores got very different probabilities when running peptide prophet). I have plotted the Kojak score vs. the peptide prophet probability for target and decoy hits and there is hardly any correlation between the kojak score and the probability.

I have repeated the search Kojak version that comes along with TPP (1.5.5) using the same settings except for the top_count parameter. I think this looks how it should, except for the cross-links which are scored very well by Kojak but get a bad peptide prophet probability. Is this the case when one peptide has a very good Kojak sub score and the other one a very bad score?

Does version 1.6.1 work for you or do you have any idea what is happening here? I attached a ppt file with the figures.

I also got a general questions: Is peptide prophet calculating FDRs separately for cross-linked, loop-linked and normal peptides?

Kind regards,

Thomas

[David Shteynberg]

Hi Thomas,

I am not sure what is happening but if you could provide the dataset files for the two search results I can investigate.   PeptideProphet does calculate different models for the different crosslink types.

Thanks,

-David

--
You received this message because you are subscribed to the Google Groups "spctools-discuss" group.
To unsubscribe from this group and stop receiving emails from it, send an email to spctools-discu...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/spctools-discuss/12f12195-f115-41e5-a196-b53e18e170df%40googlegroups.com.

[Thomas Gossenreiter]

Hi David,

thanks for your quick response. Here is the link to the zipped pep.xml files. Is this enough or do you also need the pep.xmls from the kojak search?

https://drive.google.com/open?id=1zKW0JVfVL_LME0gmt_L6AI7j8AymUmD1

Am Donnerstag, 12. September 2019 18:02:07 UTC+2 schrieb David Shteynberg:

Hi Thomas,

I am not sure what is happening but if you could provide the dataset files for the two search results I can investigate.   PeptideProphet does calculate different models for the different crosslink types.

Thanks,

-David

On Thu, Sep 12, 2019 at 8:59 AM Thomas Gossenreiter <thomas.go...@gmail.com> wrote:

Hi,

I have used Kojak 1.6.1 in the TPP pipeline and got some strange results (very similar ms2 spectra from one peptide with similar Kojak scores got very different probabilities when running peptide prophet). I have plotted the Kojak score vs. the peptide prophet probability for target and decoy hits and there is hardly any correlation between the kojak score and the probability.

I have repeated the search Kojak version that comes along with TPP (1.5.5) using the same settings except for the top_count parameter. I think this looks how it should, except for the cross-links which are scored very well by Kojak but get a bad peptide prophet probability. Is this the case when one peptide has a very good Kojak sub score and the other one a very bad score?

Does version 1.6.1 work for you or do you have any idea what is happening here? I attached a ppt file with the figures.

I also got a general questions: Is peptide prophet calculating FDRs separately for cross-linked, loop-linked and normal peptides?

Kind regards,

Thomas

--
You received this message because you are subscribed to the Google Groups "spctools-discuss" group.

To unsubscribe from this group and stop receiving emails from it, send an email to spctools...@googlegroups.com.

[David Shteynberg]

It would be very helpful to see your Kojak pepxml results.  

Thanks,

David

To unsubscribe from this group and stop receiving emails from it, send an email to spctools-discu...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/spctools-discuss/abe43fc7-eb03-4411-a6ac-c5ca4a5c326a%40googlegroups.com.

[Thomas Gossenreiter]

Please find the links to the files below:

https://drive.google.com/open?id=1bO2E0duY8gr3UCoMRClKhKlEBiLqG88D

https://drive.google.com/open?id=15BKZkfBfPsMROSvrxBOK9XtCivHG7nHk

To view this discussion on the web visit https://groups.google.com/d/msgid/spctools-discuss/abe43fc7-eb03-4411-a6ac-c5ca4a5c326a%40googlegroups.com.

[Michael Hoopmann]

Hi Thomas,

Sorry for the slow response, but HUPO is over now. I'll help take a look and find the problem.

Cheers,

Mike

To unsubscribe from this group and stop receiving emails from it, send an email to spctools-discu...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/spctools-discuss/71e9da68-1b77-4913-b2f0-049275e04709%40googlegroups.com.

[David Shteynberg]

Hi Mike and Thomas,

I am not sure where Kojak search went wrong, but somehow the 1.6.1 search contains 0 decoys ("REVERSE") tagged hits and the 1.5.5 search has many. Can you check your database and search parameters?  Also is there any reason your are not enabling the ACCMASS option that will help the classification?

Thanks,

-David

To view this discussion on the web visit https://groups.google.com/d/msgid/spctools-discuss/CANMWFMNVu2gfVxcUq_Pbttrx7WZOJHCsOtk6VOZb-H6THUtLTA%40mail.gmail.com.

[Thomas Gossenreiter]

Hi Mike and David,

first of all: thanks for your effort!

I have used the same database for both searches. If I look on the xml result file of the 1.6.1 search, I see reverse hits. The database is fine, all reverse hits start with REVERSE_

I did not enable accurate mass binning because I just wanted to try the default setting first, but if it's better for high res. data then I will switch it on in the future.

Thank you! Thomas

To view this discussion on the web visit https://groups.google.com/d/msgid/spctools-discuss/71e9da68-1b77-4913-b2f0-049275e04709%40googlegroups.com.

--
You received this message because you are subscribed to the Google Groups "spctools-discuss" group.

To unsubscribe from this group and stop receiving emails from it, send an email to spctools...@googlegroups.com.

To view this discussion on the web visit https://groups.google.com/d/msgid/spctools-discuss/CANMWFMNVu2gfVxcUq_Pbttrx7WZOJHCsOtk6VOZb-H6THUtLTA%40mail.gmail.com.