spectraCounteBayes: Peptide/Spectra Count Based Empirical Bayes Statistics for...

In yafeng/DEqMS: a tool to perform statistical analysis of differential protein expression for quantitative proteomics data.

Description

This function is to calculate peptide/PSM count adjusted t-statistics, p-values.

Usage

spectraCounteBayes(fit, fit.method="loess", coef_col)

Arguments

fit	an list object produced by Limma eBayes function, it should have one additional attribute $count, which stored the peptide or PSM count quantified for the gene in label-free or isobaric labelled data.
fit.method	the method used to fit variance against the number of peptides/PSM count quantified. Two available methods: "loess","nls" and "spline". default "loess"."loess" uses loess and span = 0.75, "nls"" uses a explicit formula y~a+b/x. "spline" uses smooth.spline and "generalized cross-validation" for smoothing parameter computation. For "nls", independent variable x is peptide/PSM count, response y is pooled variance (fit$sigma^2). For "loess" and "spline" method, both x and y are log transformed before applying the two methods. In most of time, "loess" is sufficient. To quickly assess the fit model, use VarianceScatterplot and Residualplot functions.
coef_col	an integer vector indicating the column(s) of fit$coefficients for which the function is to be performed. if not specified, all coefficients are used.

Details

This function adjusts the T-statistics and p-values for quantitative MS proteomics experiment according to the number of peptides/PSMs used for quantification. The method is similar in nature to intensity-based Bayes method (Maureen A. Sartor et al BMC Bioinformatics 2006).

Value

a list object with the following components

count	Peptide or PSM count used for quantification
sca.t	Spectra Count Adjusted posterior t-value
sca.p	Spectra Count Adjusted posterior p-value
sca.dfprior	Spectra Count Adjusted prior degrees of freedom
sca.priorvar	Spectra Count Adjusted prior variance
sca.postvar	Spectra Count Adjusted posterior variance
model	fitted model
fit.method	The method used to fit the model

Author(s)

Yafeng Zhu

Examples

library(ExperimentHub)
eh = ExperimentHub(localHub=TRUE)
query(eh, "DEqMS")
dat.psm = eh[["EH1663"]]

dat.psm.log = dat.psm
dat.psm.log[,3:12] =  log2(dat.psm[,3:12])

dat.gene.nm = medianSweeping(dat.psm.log,group_col = 2)
    
psm.count.table = as.data.frame(table(dat.psm$gene)) # generate PSM count table
rownames(psm.count.table)=psm.count.table$Var1
    
cond = c("ctrl","miR191","miR372","miR519","ctrl",
"miR372","miR519","ctrl","miR191","miR372")

sampleTable <- data.frame(
row.names = colnames(dat.psm)[3:12],
cond = as.factor(cond)
)
    
gene.matrix = as.matrix(dat.gene.nm)
design = model.matrix(~cond,sampleTable)

fit1 <- eBayes(lmFit(gene.matrix,design))
# add PSM count for each gene
fit1$count <- psm.count.table[rownames(fit1$coefficients),2]  

fit2 = spectraCounteBayes(fit1)

yafeng/DEqMS documentation built on June 3, 2020, 8:23 p.m.