Hi, Interpretation of e-values will always be coupled to the database size (more appropriately, the size of candidates for a particular spectrum). So, I do not think it is the correct metric for controlling false positives.
When the authors had suggested using e-values, I believe FDR was not in use (much). But after it was necessary to control for global error rates, Target-Decoy may be your best option. In my experience, e-values are better normalized/calibrated ( i don't know what is the right term for this) to spectrum to spectrum variations than hyperscore, it is better to use e-values than hyperscore for FDR calculation. Hyperscore have more variations and suffer from peptide length bias. Regards, *Amit Kumar Yadav * Senior Research Fellow (SRF-CSIR) IGIB, New Delhi (India) *MassWiz Web server* <http://masswiz.igib.res.in> * <http://masswiz.igib.res.in>**MassWiz sourceforge project*<https://sourceforge.net/projects/masswiz> * <https://sourceforge.net/projects/masswiz>**MassWiki*<https://sourceforge.net/apps/mediawiki/masswiz/index.php?title=MassWiki> On Thu, Mar 28, 2013 at 1:30 AM, Hanice Sun <[email protected]> wrote: > I am considering using target-decoy FDR or directly using E-value of > x!tandem? > > It seems that xtandem author recommend the latter. > > *(* > *This reference describes the idea of using reversed sequences to > validated large collections of protein identifications. The GPM has this > method built-in as a possible method for validation.* > *N.B.: We strongly recommend that you do not use this type of method for > any purpose other than comparison with other search engines. The "decoy" > search methods that have been developed from this manuscript are deeply > flawed algorithms for determining the confidence of peptide identification > assignments.* > *)* > > > However, > > 1. If I use E-value, I don't know which threshold should I set. > Although I tried to understand E-value, I still cant know which value is > suitable. > > At the beginning , I thought maybe I could set it as 1e-3, but I found the > default E-value of blast is 10, and this paper uses 100 ( > http://www.ncbi.nlm.nih.gov/pubmed/18216375). > > So I dare not set this threshold. > > 2. If I choose the target-decoy strategy, I don't know which value should > I use to sort the matching, could I use hyper-score? > > Could anyone help me? thank you very much. > > ps. My database is relatively large , It may affect the choice. > > -- > 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 http://groups.google.com/group/spctools-discuss?hl=en. > For more options, visit https://groups.google.com/groups/opt_out. > > > -- 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 http://groups.google.com/group/spctools-discuss?hl=en. For more options, visit https://groups.google.com/groups/opt_out.
