postProbNull: Posterior probabilities of each component null hypothesis

In Lululuella/OVESEG: OVESEG-test to detect tissue/cell-specific markers

Description

This function computes posterior probabilities of each component null hypothesis given observed probes. Such probe-wise probabilities will be used as weights for aggregating permutations.

Usage

postProbNull(y, group, weights = NULL, alpha = "moderated",
  detail.return = TRUE)

Arguments

y	a numeric matrix containing log-expression or logCPM (log2-counts per million) values. Data frame or SummarizedExperiment object will be internally coerced into a matrix. Rows correspond to probes and columns to samples. Missing values are not permitted.
group	categorical vector or factor giving group membership of columns of y. At least two categories need to be presented.
weights	optional numeric matrix containing prior weights for each spot.
alpha	parameter specifying within-group variance estimator to be used. 'moderated': empirical Bayes moderated variance estimator as used in eBayes. Numeric value: a constant value added to pooled variance estimator (α + σ). NULL: no estimator; all variances are set to be 1.
detail.return	a logical indicating whether more details (e.g. lfdr) will be returned.

Details

Posterior probabilities of each component null hypothesis given observed probes are estimated from ANOVA test on certain groups and local fdr. There are totally (K-1) null hypotheses, where K is the number of groups.

Value

a list containing the following components:

W	a matrix of posterior probabilities for each component null hypothesis given an observed probe. Rows correspond to probes and columns to one hypothesis.
label	a vector of group labels.
groupOrder	a matrix with each row being group indexes ordered based on decreasing expression levels. Group indexes are positions in label.
F.p.value	a matrix with each column giving p-values corresponding to F-statistics on certain groups.
lfdr	a matrix with each column being local false discovery rates estimated based on one column of weighted F.p.value matrix.
fit	a MArrayLM fitted model object produced by lmFit.

F.p.value, lfdr and fit are returned only when detail.return is TRUE.

Examples

data(RocheBT)
ppnull <- postProbNull(RocheBT$y, RocheBT$group, alpha='moderated')
ppnull <- postProbNull(RocheBT$y, RocheBT$group, alpha=0.1)

Lululuella/OVESEG documentation built on May 21, 2019, 2:28 a.m.