mehenDi

mehenDi: Tree-based differential testing using inferential uncertainty for RNA-Seq

The mehenDi method performs differential testing use the transcripts-tree obtained from TreeTerminus. The transcripts are arranged in a way such that the uncertainty decreases on ascending the tree, with the inner nodes comprising transcripts groups. mehenDi aims to leverage the tree structure to find differential signal in the tree which might get lost due to uncertainty. The selected nodes can consist of both transcripts and inner nodes, with inner nodes representing a set of transcripts that exhibit high uncertainty.

Installation

devtools::install_github("NPSDC/mehenDi")
devtools::install_github("NPSDC/brainSimSmall") ## Download the package with a small dataset

Quick start

Loading Data

dir <- system.file("extdata", package="brainSimSmall")
clustFile <- file.path(dir, "cluster_nwk.txt")
quantDir <- file.path(dir, "salmon_quants")
samples <- as.vector(outer(c(1:6), c(1,2), function(x,y) paste(x,y,sep='_')))
quantFiles <- file.path(quantDir, samples, 'quant.sf')
coldata <- data.frame(files=quantFiles, names=samples, condition=factor(rep(1:2, each=6)))

tse <- beaveR::buildTSE(treeTermFile = clustFile, coldata = coldata) ## the path to a directory that contains Salmon quantified RNA-Seq samples and the corresponding forest file produced by TreeTerminus
tse
tree <- TreeSummarizedExperiment::rowTree(tse)
tree
l <- length(tree$tip)

Running mehenDi

suppressPackageStartupMessages(library(mehenDi))
set.seed(1)
yTxps <- tse[1:l,]
yTxps <- fishpond::swish(yTxps, x="condition")

yAll <- beaveR::computeSizeFactors(tse)
yAll <- beaveR::scInfReps(yAll)
yAll <- fishpond::labelKeep(yAll)
metadata(yAll)$infRepsScaled <- TRUE
yInn <- yAll[(l+1):nrow(yAll),]
set.seed(1)
yInn <- fishpond::swish(yInn, x="condition")

pvals <- c(mcols(yTxps)[["pvalue"]], mcols(yInn)[["pvalue"]])
tD <- mehenDi::mehenDi(yAll, x="condition", pvalues = pvals,
                    minP=0.7, mIRVThresh=0.4, alpha=0.01)
length(tD$selNodes) ## Total number of selected nodes
sum(tD$selNodes > l) ## Number of inner nodes that are selected

The input to mehenDi is the TreeSummarizedExperiment (tse) object. This object can be created using the package beaveR, which requires the same input as tximeta and the forest file produced by TreeTerminus. Among the other inputs are the p-values corresponding to each node in tree belonging to the tse object, which here in this example have been generated using swish in the package fishpond.

Exploration of the nodes output by mehenDi

fishpond::plotInfReps(yAll, tD[["selNodes]][1], x="condition")
nodeInf <- beaveR::findNodeInformation(yAll, node = tD[["selNodes]][1], type = "tips")
nodeInf