Generate statistics associated with pairwise differential expression from RNAseq count data

Description

When provided with an CountDataSet, comparisons are made between control and perturbation samples.

Usage

1
2
 pairwise_DESeq(cds, vst, control_perturb_col = "condition",
control="control", perturb="perturbation", try.hard=FALSE) 

Arguments

cds

CountDataSet with all count data for a single instance, plus metadata on which samples are perturbation and control.

vst

Matrix of variance-stabilized count data that must include columns with colnames matching the sampleNames of the cds object. The vst matrix may contain additional columns / samples, which will be ignored.

control_perturb_col

Column name in phenoData of cds where control/perturbation designations are stored.

control

String designating control samples in the control_perturb_col column.

perturb

String designating perturbation samples in the control_perturb_col column.

try.hard

Logical parameter indicating how to proceed when DESeq's parametric estimation of the dispersion parameter fails. If set to FALSE (default), the function exits with an error. If set to TRUE, the function will try a non-parametric approach instead.

Value

The function returns a data frame with the following columns:

log_fc

Moderated log2 fold-change between perturbed and control data. (A positive value denotes higher expression in the perturbed samples.) The change was calculated from the (mean) counts after variance stabilizing transformation. Please consult the DESeq vignette for details on the transformation.

z

For ease of comparison across instances with different numbers of samples, the (uncorrected) DESEq p-value is converted to the standard normal scale. The result is reported here. As for log_fc, positive values denote higher expression in perturbed samples.

p

p-value for differential expression calculated by the nbinomTest function from the DESeq package. In the absence of replicates, the dispersion parameter is estimated across all samples, ignoring the class labels, by using the blind method of the estimateDispersions function. When replicates are available, the pooled method is used instead. Note that this p-value can also be computed from z, via pnorm (doubling for two tails).

Note

To use this function, please install the suggested Bioconductor package 'DESeq'.

Want to suggest features or report bugs for rdrr.io? Use the GitHub issue tracker.