Statistics and Its Interface

Volume 5 (2012)

Number 1

Generalized linear and mixed models for label-free shotgun proteomics

Pages: 89 – 98

DOI: https://dx.doi.org/10.4310/SII.2012.v5.n1.a8

Authors

Matthew C. Leitch (Sealy Center for Molecular Medicine, The University of Texas Medical Branch, Galveston, Texas, U.S.A.)

Indranil Mitra (Sealy Center for Molecular Medicine, The University of Texas Medical Branch, Galveston, Texas, U.S.A.)

Rovshan G. Sadygov (Sealy Center for Molecular Medicine, The University of Texas Medical Branch, Galveston, Texas, U.S.A.)

Abstract

Label-free shotgun proteomics holds great promise, and has already had some great successes in pinpointing which proteins are up or down regulated in certain disease states. However, there are still some pressing issues concerning the statistical analysis of label-free shotgun proteomics, and this field has not enjoyed as much dedication of statistical research towards it as microarray research has. Here we reapply previously used statistical methods, the QSpec and quasi-Poisson, as well as apply the negative binomial distribution to both a control data set and a data set with known differential expression to determine the successes and failure of each of the three methods.

Keywords

count data, statistical models, spectral count, label-free quantitative proteomics, p-values, FDR, negative binomial model, quasi-Poisson model, mixture model, generalized linear models

Published 17 February 2012