PTMProphet and best localization score

[Niko Pinter]

Hi,

thanks for all the effort you put into TPP and PTMProphet.

We are currently analyzing some datasets of synthetic phosphopeptides with PTMProphet and wondered where to set the cut-off for the best localization score. A score above 0.750 was often used in literature and we tend to use a similar cut-off.

Best regards
Niko Pinter

[David Shteynberg]

Hi Niko,

Thanks for your question and interest in the TPP.

The PepXMLViewer.cgi tool can be used to display, filter and export pepXML data, including PTMProphet processed results.  PTMProphet computes several statistics for each PSM and for each PTM analyzed:  

Mean Best Probability -- is the average PTMProphet probability of the top m modified sites of a given type in the peptide

Normalized Information Gain -- is the information gain based on the PTMProphet probabiliy of the top m modified sites of a given type in the peptide

Localized Modification Count -- is the expected number of correct modifications of a given type localized with certainty on this peptide

For more information about there metrics please refer to the PTMProphet publication.

The reason you might want to use the more conservative Info Gain metric to threshold your results is that site probabilities are not directly comparable between PSMs with different site and PTM counts, but Info Gains are directly comparable.  E.g. a Mean Best PTM probability of 0.75 on a peptide with 4 sites and 3 PTMs of the a given type means there is zero information about the location of those 3 PTMs and it means a different thing from a  Mean Best PTM probability of 0.75 on a peptide with 4 sites and 1 PTM where the information regarding the location of the PTM is greater than zero!

You can set thresholds for any of these metrics on the Filter page in PepXMLViewer.cgi. Make sure to set it for the correct PTM type.

I hope I have understood your question correctly, please let me know if you have any more.

Cheers!

-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/fe600cfc-5728-4f83-b564-13ba9a33b4a7n%40googlegroups.com.