Hi Pete, Thank you for the summary. I had a question about the 1% FDR. Is this decoy-based or model-based? I am wondering what protein counts you will observe when you compare at the 1% decoy-based FDR between running NSP in both iProphet and ProteinProphet, running NSP only in iProphet and running NSP only in ProteinProphet?
Thank you, -David On Thu, Jun 21, 2018 at 10:39 AM, <[email protected]> wrote: > Hi David, > > Thanks a lot for your reply. > > Your first answer makes a big difference - I had been combining iprophet > results for a second round which had previously combined multiple search > engines. By combining the results from different search engines only in the > second iProphet run, my numbers are now more consistent irrespective of > order of combination. > > Regarding the second point, switching NSP off in protein prophet (after > running NSP in iProphet) makes quite a big difference to my final protein > numbers (2632 entries versus 2336 at 1% FDR). > > Taking a look at the additional matches, they are all single peptide > sequence hits - however, from manual inspection of several of these though, > there are often multiple matches with high peptide prophet scores across > several different biological replicates, and the spectra look good. Some of > them do look like single matches though (albeit with good looking spectra). > > Given that both are 1% FDR, it's difficult to choose the most appropriate > method to choose. From what I've said, I think the risk of false negatives > is greater (running both NSP models) than the risk of false positives > (running iprophet NSP only) - my thoughts are that it will be better to use > NSP 'off' in protein prophet, but to consider protein IDs from single > peptide sequence hits as being less confident. > > Thanks again for your help, > Pete > > > > > On Wednesday, 20 June 2018 13:29:06 UTC-7, David Shteynberg wrote: >> >> Hello Pete, >> >> I think the answer to your first question is it depends on the specifics >> of your analysis. >> >> You can pass iProphet files through iProphet again, since it will just >> use the PeptideProphet probabilities which are not modified (only reported) >> by iProphet. If the iProphet is from a single search engine this should be >> just fine. However, if the iProphet file contains results from multiple >> search engines then you probably don't want to combine it with iProphet >> again as in this case each spectrum search result comes from the highest >> scoring search engine, so not all the information will be available for >> iProphet in the second analysis. Also, for your large analysis that is >> currently failing in Petunia you might consider running the tool on the >> commandline. >> >> For your second question, the NSP model can be disabled on the >> commandline using the NONSP flag. The ProteinProphet NSP model is >> implemented differently in iProphet and in ProteinProphet. Although, in >> theory applying the models both times could be problematic due to their >> similarity, in practice the models are different enought that during >> testing I have not observed deleterious effects from using both the >> iProphet and the ProteinProphet NSP models. You can try running the tools >> in different ways and comparing the performance. Using both NSP models is >> the current default and you would have to explicitly disable the models >> when you run each tool. >> >> I hope this helps. >> >> -David >> >> >> >> >> On Tue, Jun 19, 2018 at 5:29 PM, <[email protected]> wrote: >> >>> Hi All, >>> >>> I have 2 questions I'd be grateful if people could help answer: >>> >>> 1) >>> Is it valid to combine multiple iprophet.pep.xml files by passing >>> through iprophet for a second time? - alternatively, is it valid to combine >>> a single iprophet.pep.xml file with interact.pep.xml files in iprophet? >>> >>> I am trying to combine a lot of different experiments / search engines >>> results etc, and have been combining in iprophet - but I appear to have >>> maxed out the number interact.pep.xml files to pass into iprophet. Beyond a >>> certain number of files (doesn't appear to be file-specific), iProphet >>> fails. As a workaround, I wondered if I could simply run half of the files >>> through iProphet, then combine the resulting file with the remaining files >>> to be run, by running iprophet again prior to running protein prophet. - >>> Would this be valid? >>> >>> 2) >>> I attended a TPP course last year in which the course notes stated that >>> NSP should be switched off in iProphet, if NSP model is to be used in >>> protein prophet. >>> >>> I am using petunia (running protein prophet on the iprophet results), >>> and I cannot see a NSP option in the protein prophet parameters. Does this >>> mean that NSP is not being used when I run protein prophet? ... (i.e. am I >>> ok to leave NSP on in iProphet?) >>> >>> >>> 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 [email protected]. >>> To post to this group, send email to [email protected]. >>> Visit this group at https://groups.google.com/group/spctools-discuss. >>> For more options, visit https://groups.google.com/d/optout. >>> >> >> -- > 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 [email protected]. > To post to this group, send email to [email protected]. > Visit this group at https://groups.google.com/group/spctools-discuss. > For more options, visit https://groups.google.com/d/optout. > -- 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 [email protected]. To post to this group, send email to [email protected]. Visit this group at https://groups.google.com/group/spctools-discuss. For more options, visit https://groups.google.com/d/optout.
