decoy detection

[Kris]

Hi Everyone,

I am trying to detect false positive rates in my TPP runs. I want to
make sure that my probability result measures are actually correct. I
have read some papers describing high levels of false positives and
using decoy sets to correct for these errors.

I originally downloaded Mayu, made a fasta file with the appended
decoy set (appending rev to the beginning of decoy sequences). I ran
this through SEQUEST (inside of TPP). I tried to run Mayu (through
command line) on the pepxml results of TPP. But I kept getting an
error where it said the input file has no data. I think it was not
parsing the pep.xml file correctly.

Then i noticed that TPP has an option for running Peptide Prophet with
a decoy correction. I took my output and ran peptide prophet with the
decoy correction enabled. I noticed the output has decreased
probability scores for the lower probability scores when this
correction was enabled.


MY QUESTION:

What is the "Use decoy hits to pin down the negative distribution"
checkbox actually doing? Is it using the Mayu software? And if it is
not using Mayu, is it as accurate at reducing false positives as
Mayu?

Is there any change in the output other than probability score?

Is there anything I need to know or do differently to correctly filter
out false positives? I would like to have increased confidence in my
peptide matches because our group is trying to match new sequences
that you would not normally find in the mass spec data.


Thanks for all the help,
Kris

[Kris]

I feel like my earlier post may have been too long and intimidating
for people to respond to. So, I'm hoping this concise addition will
get a response.

Does the internal TPP "Use decoy hits to pin down the negative
distribution" checkbox (in the Analyze Peptides section) use Mayu for
its analysis?

If not, what's different between Mayu and the TPP version?

Lastly, which decoy detection probably works better (which should I be
using)? and why?

I'll be checking the forums for responses.

Thanks,
Kris

[Jimmy Eng]

In a gross simplification, PeptideProphet fits positive and negative
distributions to the overall search results distribution as part of
its processing to calculate probability scores for each
peptide-to-spectrum match. That checkbox you're referring to tells
PeptideProphet to use the decoy matches to model the negative
distribution because the decoy hits are explicitly known negatives.
That option is possibly helpful but not needed; in fact you do not
have to even need to do a decoy search to use PeptideProphet.

Mayu is a completely separate tool that calculates protein level false
discovery rates (as well as peptide level and peptide-to-spectrum
match level FDR). FWIW, I personally haven't used Mayu so I can't
help you with it but others here hopefully can.

But the real question on what tool(s) you should be using is based on
what questions you're asking of your data. Have you looked at
Peptide/ProteinProphet output and compared that to Mayu output to see
how useful the results from each are to you? What's your criteria for
better?