A Nonlinear Scoring Framework for Peptide Identification via Tandem Mass Spectrometry

Yan Fu,Qiang Yang,R. Sun,C. Ling,Dequan Li,Hu Zhou,Simin He,Wen Gao
2004-01-01
Abstract:The problem of false positives in peptide identification via tandem mass spectrometry (MS/MS) by database searching remains unsatisfactorily resolved in the current proteomics research. The correlative information among fragment ions in the MS/MS spectrum can be very helpful for reducing the number of false positives. However, due to the computational difficulty, existing peptide-scoring algorithms usually assume the independence of the occurrences of fragment ions and employ linear scoring functions. In our earlier work, we have developed a nonlinear peptide-scoring function called Kernel Spectral Dot Product (KSDP, see Fu et al., 2004), in which co-occurring matches of consecutive fragment ions are emphasized by the locally improved polynomial kernel. In this paper, we extend the KSDP to a general framework that can readily accommodate different kernel functions. We show that by using the Radial Basis Function (RBF) kernel for consecutive fragment ions, ∗ To whom correspondence should be addressed. better performance can be achieved. Experiments on the KSDP for complementary and homologous fragment ions reveal that these two kinds of correlation are not necessarily useful for reducing false positives. Our software tool, pFind, obtains higher identification accuracy on a previously reported dataset than two popular software tools, SEQUEST and Sonar MS/MS.
What problem does this paper attempt to address?