How do I calculate a false discovery rate (FDR) for my set of identifications by spectral library searching?
You may use a “target-decoy” approach similar to what is used for sequence searching (Elias and Gygi, Nat. Meth., Mar. 2007). This has been demonstrated by generating decoy spectra for spectral library searches by Lam et al, J. Prot. Res., Jan. 2010. Additionally, any set of non-overlapping spectra (e.g, from another organism) may be used as decoy by adjusting for any “target-decoy” bias. To generate the figure below, a set of human spectra were searched against a combined library of human spectra and non-overlapping, non-human spectra chosen at random from other libraries. The black bars represent the fraction of matches to the human spectra and the red to decoy spectra. At ranks >3 for any set of matches, the matches are expected to be near random. Therefore, any deviation from 50:50 can be described as a “target-decoy” bias. In the example below the bias factor would be roughly 62/38 or 1.6 in favor of the target spectra. This value can then be used to scale the number of decoy (fal
