rnaseqcomp (version 1.2.2)

plotROC: Estimate And Plot Differential Expression

Description

For each pipeline, differential expression is first estimated by fold change on 1 vs. 1 comparison between cell lines. ROC curves then are made by comparing fold changes with predefined true differentials. Then, ROC curves from multiple 1 vs. 1 comparisons are averaged using threshold averaging strategy. Standardized partial area under the curve (pAUC) is reported for each pipeline.

Usage

plotROC(dat, positive, fcsign, cut = 1, constant = 0.5,
  thresholds = seq(12, 0, len = 300), arrow = FALSE, ...)

Arguments

dat

A rnaseqcomp S4 class object.

positive

A logical vector with length equivalent to row number of matrices in dat@quantData. TRUE means true differential and FALSE means true non-differential, while missing value NA means unknown.

fcsign

A numeric vector with length equivalent to row number of matrices in dat@quantData. Only values {1, -1, 0} are allowed. 1 means upregulated in second cell line, -1 means downregulated in second cell line, and 0 means no change. If elements in fcsign correspond to NA in positive, these elements will be ignored in estimation.

cut

A numeric cutoff used to decide if fold change should be estimated. For a 1 vs 1 comparison, if features have signals less than cut in both samples, their fold changes will be set to 0. (default: 1)

constant

A numeric constant that is added to quantifications before fold changes calculation. (default: 0.5)

thresholds

A numeric vector defining cutoffs on fold changes as the points to make threshold averaging on ROC curves. (default: seq(12, 0, len = 300))

arrow

A logical indicating if error bars should be added to the averaged ROC curves. (default: FALSE)

...

Parameters for base function plot.

Value

• plot
  ROC plots for all the quantification pipelines.
• pAUC
  A numeric vector indicating pipeline accuracy. This is standardized partial AUC based on ranges chosen on false positive rate.

Examples

data(simdata)
condInfo <- factor(simdata$samp$condition)
repInfo <- factor(simdata$samp$replicate)
evaluationFeature <- rep(TRUE, nrow(simdata$meta))
calibrationFeature <- simdata$meta$house & simdata$meta$chr == 'chr1'
unitReference <- 1
dat <- signalCalibrate(simdata$quant, condInfo, repInfo, evaluationFeature,
calibrationFeature, unitReference, calibrationFeature2 = calibrationFeature)
plotROC(dat,simdata$meta$positive,simdata$meta$fcsign)