Peptide prophet - use of decoy database for X! searches?

[Peter Bell]

Please can someone advise - is it recommended to include decoys in a database when using X! for searches in TPP?

As I understand, X! recommends not including decoys in searches (when used standalone) – however when I analyse X! identified peptides using peptide/protein prophet, I assume that I will need decoy matches to help pin down the distribution?

Thanks

Pete

[David Shteynberg]

Hello Pete,

Use of decoys in the X!Tandem workflow is not strictly required but it is helpful.  X!Tandem is one of a handful of search engines for which there exists a parametric mixture model for PSM-level validation.  In my experience, using the parametric models may not always give you the best results in terms of validation, and the lack of decoys precludes the ability to compute decoy-based error rates (FDR).  In this case you have to only the FDR as estimated by the PeptideProphet model, and you cannot compare this FDR to decoy-based estimates.  The best results are usually obtained by using a database that includes some proportion of decoys.  You can then validate the search results using PeptideProphet's semi-supervised semi-parametric models, which tend to estimate more accurate error rates.  However you decide to search your data, I also suggest you also include iProphet in your validation analysis for peptide-level validation.

Cheers,

-David

On Tue, Jun 5, 2018 at 12:47 PM, <pbell....@gmail.com> wrote:

Please can someone advise - is it recommended to include decoys in my database when using X! for searches then peptide/protein prophet in TPP?

As I understand, X! recommends not including decoys in searches (when used standalone) – however when I analyse X! identified peptides using peptide/protein prophet, I assume that I will need decoy matches to help pin down the correct distribution?

Thanks

Pete

--
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-discuss+unsubscribe@googlegroups.com.
To post to this group, send email to spctools-discuss@googlegroups.com.
Visit this group at https://groups.google.com/group/spctools-discuss.
For more options, visit https://groups.google.com/d/optout.

[pbell....@gmail.com]

Hi David,

Thanks for your reply, that's very helpful. 

I've now tried including an equal number of decoys (reversed) in my database, processing the same data through peptide prophet, iprophet and protein prophet. My number of identified proteins has gone down, but the PSMs are much more believable (judging from the spectra).

As an aside - if I were to play with the proportion of decoys in the database, (reducing the number to say 25% of the total database, rather than 50%), in your experience would this reduction in the search space be likely to have much of an effect on PSM / protein IDs?

Thanks again,

Pete

On Tuesday, 5 June 2018 16:26:37 UTC-7, David Shteynberg wrote:

Hello Pete,

Use of decoys in the X!Tandem workflow is not strictly required but it is helpful.  X!Tandem is one of a handful of search engines for which there exists a parametric mixture model for PSM-level validation.  In my experience, using the parametric models may not always give you the best results in terms of validation, and the lack of decoys precludes the ability to compute decoy-based error rates (FDR).  In this case you have to only the FDR as estimated by the PeptideProphet model, and you cannot compare this FDR to decoy-based estimates.  The best results are usually obtained by using a database that includes some proportion of decoys.  You can then validate the search results using PeptideProphet's semi-supervised semi-parametric models, which tend to estimate more accurate error rates.  However you decide to search your data, I also suggest you also include iProphet in your validation analysis for peptide-level validation.

Cheers,

-David

On Tue, Jun 5, 2018 at 12:47 PM, <pbell....@gmail.com> wrote:

Please can someone advise - is it recommended to include decoys in my database when using X! for searches then peptide/protein prophet in TPP?

As I understand, X! recommends not including decoys in searches (when used standalone) – however when I analyse X! identified peptides using peptide/protein prophet, I assume that I will need decoy matches to help pin down the correct distribution?

Thanks

Pete

--
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 post to this group, send email to spctools...@googlegroups.com.

[David Shteynberg]

It seems like a reasonable thing to test.  I am not sure of the ultimate result but maybe you'll be able to make a few more identifications this way.  Let me know what happens if you end up trying that.

-David

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