Analysis of DIGE data using a linear mixed model allowing for protein‐specific dye effects

Abstract

Differential in-gel electrophoresis (DIGE) experiments allow three protein samples to be run per gel. The three samples are labeled with the spectrally resolvable fluorescent dyes, Cy2, Cy3, and Cy5, respectively. Here, we show that protein-specific dye effects exist, and we present a linear mixed model for analysis of DIGE data which takes dye effects into account. A Java implementation of the model, called DIGEanalyzer, is freely available at http://bioinfo.thep.lu.se/ digeanalyzer.html. Three DIGE experiments from our laboratory, with 173, 64, and 24 gels, respectively, were used to quantify and verify the dye effects. DeCyder 5.0 and 6.5 were used for spot detection and matching. The fractions of proteins with a statistically significant (0.001 level) dye effect were 19, 34, and 23%, respectively. The fractions of proteins with a dye effect above 1.4-fold change were 1, 4, and 6%, respectively. The median magnitude of the dye effect was 1.07-fold change for Cy5 versus Cy3 and 1.16-fold change for Cy3 versus Cy2. The maximal dye effect was a seven-fold change. The dye effects of spots corresponding to the same protein tend to be similar within each of the three experiments, and to a smaller degree across experiments.

Plasma Proteomic Analysis of Simian Immunodeficiency Virus Infection of Rhesus Macaques

“…Typically two quantitative proteomic platforms are used for profiling plasma: gel based 2-dimensional electrophoresis usually with difference gel electrophoresis (2DE DIGE) 20–22 and mass spectrometry based quantitation using isobaric Tags for Absolute and Relative Quantitation (iTRAQ) 23, 24 . In our previous investigations of plasma samples we used 2DE DIGE platform which resulted in identification of previously not reported differential expression of proteins such as afamin and gelsolin 18, 25, 26 .…”

Image analysis tools and emerging algorithms for expression proteomics

“…Techniques that consider linear fixed and random effects are termed ‘linear mixed models’. Two‐stage hierarchical linear mixed models have been applied to 2‐D DIGE spot lists for normalisation of protein‐specific dye effects 124, 125. For SELDI MS, Handley 112 combined parametric mixture modelling with a two‐level linear mixed model.…”

2D DIGE Data Analysis for Differential Protein Spot Identification

“…Image pairs are then matched between gels using Decyder-BVA software, which looks for consistency of differences between samples across all the gels and applies student t-test based statistics to associate a level of confidence for each of those differences. Due to the fact that there will be a very small sample size or replications in most experiments, the inherent assumption of normal distribution of data is usually hard to meet for t-test, which may result in a misleading result in identifying the differentially expressed protein spot by using the p-value of test [6]. In fact, based on our preliminary study, small p value and reasonable volume ratio cannot guarantee the spot to be a significant DE protein spot.…”