We have been discussing this in our group and I was wondering whether anyone knows a reference/publication about this:
Everyone uses different search engines (X!Tandem, SEQUEST, MASCOT, etc), in combination with Peptide/ProteinProphets lets say. When looking at the intersection of the lists of peptide/protein identifications from these searches, people usually find some overlap, but different pipelines seem to identify different peptides from the same RAW. One could imagine simply combining (taking the union of) the peptide/protein lists that are identified in the different approaches if one is shooting for high coverage.
Is there any work on a statistically sound combination of different techniques, something like a 'meta-search' for mass spectrometry data? Something that combines different search engines into a common framework? Something beyond use of the union or intersection of individual protein or peptide lists? Is there a study in which someone would simply use the Peptide/ProteinProphets on PEPXML files that come from, say, X!Tandem and SEQUEST run on the same RAW file?
I work on a project called Scaffold that does "meta-searching" with
multiple search engines, among other things. The statistical method
we use is based on the Peptide and ProteinProphet algorithms. Alexey
Nesvizhskii and I wrote a paper about this method in JPR called
"Improving sensitivity by probabilistically combining results from
multiple MS/MS search methodologies" (J Proteome Res. 2008 Jan;7(1):
245-53). Please contact me offline if you have questions about how
the method works!
> We have been discussing this in our group and I was wondering whether anyone
> knows a reference/publication about this:
> Everyone uses different search engines (X!Tandem, SEQUEST, MASCOT, etc), in
> combination with Peptide/ProteinProphets lets say. When looking at the
> intersection of the lists of peptide/protein identifications from these
> searches, people usually find some overlap, but different pipelines seem to
> identify different peptides from the same RAW. One could imagine simply
> combining (taking the union of) the peptide/protein lists that are
> identified in the different approaches if one is shooting for high
> coverage.
> Is there any work on a statistically sound combination of different
> techniques, something like a 'meta-search' for mass spectrometry data?
> Something that combines different search engines into a common framework?
> Something beyond use of the union or intersection of individual protein or
> peptide lists?
> Is there a study in which someone would simply use the
> Peptide/ProteinProphets on PEPXML files that come from, say, X!Tandem and
> SEQUEST run on the same RAW file?
We do have a tool in the TPP, called iProphet, for combining search
engines. This tool applies a set of 5 statistical models of which the
search engine combining is merely one. We will have a paper coming
out on iProphet in the near future. Meanwhile, the tool is available
to try in the TPP. Please check the following thread for some further
instructions:
On Fri, Jun 26, 2009 at 5:45 PM, Brian Searle<sear...@gmail.com> wrote:
> Hi Christine,
> I work on a project called Scaffold that does "meta-searching" with
> multiple search engines, among other things. The statistical method
> we use is based on the Peptide and ProteinProphet algorithms. Alexey
> Nesvizhskii and I wrote a paper about this method in JPR called
> "Improving sensitivity by probabilistically combining results from
> multiple MS/MS search methodologies" (J Proteome Res. 2008 Jan;7(1):
> 245-53). Please contact me offline if you have questions about how
> the method works!
>> We have been discussing this in our group and I was wondering whether anyone
>> knows a reference/publication about this:
>> Everyone uses different search engines (X!Tandem, SEQUEST, MASCOT, etc), in
>> combination with Peptide/ProteinProphets lets say. When looking at the
>> intersection of the lists of peptide/protein identifications from these
>> searches, people usually find some overlap, but different pipelines seem to
>> identify different peptides from the same RAW. One could imagine simply
>> combining (taking the union of) the peptide/protein lists that are
>> identified in the different approaches if one is shooting for high
>> coverage.
>> Is there any work on a statistically sound combination of different
>> techniques, something like a 'meta-search' for mass spectrometry data?
>> Something that combines different search engines into a common framework?
>> Something beyond use of the union or intersection of individual protein or
>> peptide lists?
>> Is there a study in which someone would simply use the
>> Peptide/ProteinProphets on PEPXML files that come from, say, X!Tandem and
>> SEQUEST run on the same RAW file?
|
