In soderling-lab/tidyProt: tidy statistical analysis of protein mass spectrometry

edgeR Goodness of Fit for TMT Protein MS

Most analysis tools for proteomics data have been borrowed directly from packages designed explicitly for transcriptomics or RNA-seq data analysis: these approaches may make assumptions that may not hold for proteomics data and should be carefully considered.

For example, the edgeR package fits the data with a generalized linear model which assumes that the data are well described by a negative binomal model: these may be appropriate RNA sequencing-based cound data, but not for proteomics data. The underlying reason for difference in statistical nature of the data may stem from the difference in RNA and protein abundance in the cell.

The overall adequacy of edgeR's NB GLM for TMT data were assessed by plotting the residual deviance for all proteins as a quantile-quantile plot (McCarthy2012). (A) Normalized protein-level data were fit with a NB GLM of the form:
log2(Intensity) ~ Mixture + Condition.

Where Mixture is an additive blocking factor that accounts for variability between TMT mixtures. The NB framework used by edgeR utilizes a dispersion parameter, psi, to model mean-variance relationships in the data in three different ways: 'common', 'trended', and 'tagwise'. I plot the deviance statistics for the data fit with each of the three dispersion parameters against their theoretical normal quantiles using the edgeR::gof function. (B) The normalized protein-level data were fit with a linear mixed-effects model using lme4::lmer (Bates2014):
log2(Intensity) ~ 1 + Condition + (1|Mixture).

Where Mixture represents the mixed-effect of Mixture. The residual deviance and degrees of freedom were extracted from the fitted models, z-score normalized, and plotted as in (A). Proteins with a significantly poor fit are indicated as outliers in blue (Holm-adjusted P-value < 0.05).

Session info {.unnumbered}

This document was compiled with rmarkdown::render().

sessionInfo()

soderling-lab/tidyProt documentation built on Dec. 23, 2021, 3:31 a.m.