Improved Peptide Identification Using Implicit Properties Learned From Transductive Support Vector Machines

摘要

Based on the above results, we can conclude that this method holds promise for increasing the number of PSMs positively identified at the same significance level in a tandem MS/MS database search. In IEF fraction 2, we see that our method is approaching the estimated limit of the number of true PSMs in the database search results. In IEF fraction 10, we see much more modest gains in the number of PSMs identified at a 5percent FDR. In IEF fraction 18, the fraction used for optimization of the SVM parameters, we see significant gains over using the Mascot ion score alone to separate true PSMs from false PSMs. One explanation for relatively poor performance on IEF fraction 10 is that there was a wider range of assumed true PSM pIs in IEF fraction 10 (5.2-8.8) compared to IEF fraction 2 (4.8-6.1) and IEF fraction 18 (6.7-8.8), thus making it harder to discriminate true PSMs from false PSMs in IEF fraction 10.