In response to "On E-value for tandem MS scoring schemes"

Abstract

We thank Mark Segal for raising the issue of interpreting MS/MS scores. As he noted, we used a method proposed by Fenyo and Beavis (FB) (2003) to asses the significance of identification using HMM_Score. In his letter, Segal makes two basic assertions about this use: (1) that the extreme value distribution does not apply for the MS/MS database scoring systems used by FB and our HMM and (2) the linear tail fitting of the log survival function is not robust. He proposes a method that he authored as an alternative for estimating evd parameters that he says may be more robust, and also points to a method by Shen et al. that is specific to assessing significance of proteins/peptides identifications using MS/MS data. While it is valuable to examine whether there exist better ways of statistically interpreting the results of MS/MS search, in his letter, Segal did not provide any clear supporting evidence for his claim that the MS/MS scorers cannot use E-values. In our case, we calculate a score distribution for all random matches on-the-fly, then deriving the survival function, s, (the cumulative probability distribution) and finally, fitting a line to log of this function for the high-scoring portion of s. We verified the methodology for a series of randomly chosen HMM_Score search results, observing that in all cases, the fit had very high correlation values (R2 > 0.9). All subsequent validation of HMM_Score was performed using the E-values produced, and as reported the system performs well.