Kojak + PeptideProphet FDR cutoff

[Thomas Gossenreiter]

Dear all,

I would like to filter my Kojak cross-linking data at a defined FDR. Actually I would like to apply a separate FDR for loop-links, cross-links and non-linked peptides.

So I am wondering if I can use the Error Table provided by the PeptideProphet results to get the minimum probability to apply a certain FDR on my dataset? Or doesn't this work for a dataset containing mixed peptide species?

Best regards,

Thomas

[David Shteynberg]

Hi Thomas,

The tables for each type of linked peptide is in the pepXML file (near the top of the file), listed separately for each type of link.   

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/f3dffce4-ac1a-4512-92af-9ef459e8bbd3%40googlegroups.com.

[Thomas Gossenreiter]

Thank you David, I found it!

[Thomas Gossenreiter]

Dear David,

is it correct that the FDR in PP using the target-decoy approach is calculated with this formula:

FDR = decoy/target = false positives / (false positives + true positives) ?

Best regards,

Thomas

[David Shteynberg]

No it is not correct for xl peptides.  There are two peptides in every match thus id you assume the database contains half decoys and half targets, then there are 3/4 chance to match a decoy peptide randomly and only 1/4 chance to match a target.  So the random xl matches will be 3/4 decoy and 1/4 target so your decoy based error rate should estimate the number of wrong hits by taking the number of decoy matches and multiplying by 4/3.  So wrong = decoy*4/3 and errorrate = wrong/total.

Hopefully this makes sense.

David 

[Thomas Gossenreiter]

Okay thank you, so PP calculates the FDR of cross-linked peptides like that?

((number of DD+TD+DT)*(4/3))/number of TT ?