msmsTests (version 1.10.0)

test.results: Multitest p-value adjustment and post-test filter

Description

Operates on the statistic tests results obtained from msms.glm.pois(), msms.glm.qlll() or msms.edgeR(). The following variables are computed: Raw expression mean values for each condition (control and treatment), log fold change based on these expression levels and taking into account the normalizing divisors (div), multitest adjusted p-values with FDR control, and a post test filter based on minimum spectral counts and minimum absolute log fold change as estimated by the statistic test. According to the results of this post-test filter, features are flagged as T or F depending on whether they result relevant or not, beyond their statistic signicance.

Usage

test.results(test, msnset, gpf, gp1, gp2, div, alpha = 0.05,  minSpC = 2, minLFC = 1, method = "BH")

Arguments

test

The dataframe obtained from either msms.glm.pois(), msms.glm.qlll() or msms.edgeR()

msnset

A MSnSet object with spectral counts in the expression matrix.

gpf

The factor used in the tests.

gp1

The treatment level name.

gp2

The control level name. Should be the factor's reference level. See R function relevel.

div

The weights used as divisors (offsets) in the GLM model. Usually the sum of spectral counts of each sample.

alpha

The multi test adjusted p-value significance threshold.

minSpC

The minimum spectral counts considered as relevant in the most abundant condition. This filter aims at reaching good reproducibility.

minLFC

The minimum absolute log fold change considered both, relevant and biologically significant. This filter aims at assuring enough biological effect size and at reaching good reproducibility.

method

One among BH or qval. The p-values are FDR ajusted by the Benjamini-Hochberg method (BH) or by qvalue (qval).

Value

A data frame with the following columns:

first column

Column named as the treatment level with the mean raw spectral counts observed for this condition

second column

Column named as the control level with the mean raw spectral counts observed for this condition

lFC.Av

Log fold change computed from the mean expression levesl taking into account the given normalization factors.

logFC

Log fold change estimated by fitting the given GLM model. The reference level of the main factor is taken as control.

D or LR

The statistic obtained from the tests. The residual deviance D for Poisson and quasi-likelihood, or the likelihood ratio LR for edgeR.

p.val

The unadjusted p-values obtained from the tests.

adjp

The multitest adjusted p-values with FDR control.

DEP

A logical flagging the features considered both as statistically significant and relevant for reproducibility.

Details

No feature is removed in the filter, but instead they are flagged as TRUE or FALSE depending on whether they are considered as differentially expressed or not, in the DEP column, taking into account statistic significance and reproducibility metrics.

References

Josep Gregori, Laura Villareal, Alex Sanchez, Jose Baselga, Josep Villanueva (2013). An Effect Size Filter Improves the Reproducibility in Spectral Counting-based Comparative Proteomics. Journal of Proteomics, DOI http://dx.doi.org/10.1016/j.jprot.2013.05.030 Benjamini, Y., and Hochberg, Y. (1995). Controlling the false discovery rate: a practical and powerful approach to multiple testing. Journal of the Royal Statistical Society Series B, 57, 289-300.

Alan Dabney, John D. Storey and with assistance from Gregory R. Warnes. qvalue: Q-value estimation for false discovery rate control. R package version 1.30.0.

See Also

pval.by.fc, p.adjust, qvalue,

Examples

library(msmsTests)
data(msms.dataset)
# Pre-process expression matrix
e <- pp.msms.data(msms.dataset)
# Factors
pData(e)
# Control condition
levels(pData(e)$treat)[1]
# Treatment condition
levels(pData(e)$treat)[2]
# Models and normalizing condition
null.f <- "y~batch"
alt.f <- "y~treat+batch"
div <- apply(exprs(e),2,sum)
#Test
res <- msms.glm.qlll(e,alt.f,null.f,div=div)
# Post-test filter
lst <- test.results(res,e,pData(e)$treat,"U600","U200",div,
                    alpha=0.05,minSpC=2,minLFC=1,
                    method="BH")
str(lst)
lst$cond
head(lst$tres)
rownames(lst$tres)[which(lst$tres$DEP)]