CNsolidate

CNV detection using 12 independent change point detection algorithms and an expert voting system

See example at the bottom for complete walkthrough using synthetic data

INSTALL

Download zip file or pull master branch

R CMD build CNsolidate-master
R CMD install CNsolidate_1.2.tar.gz

Load library with R in the usual way:

library(CNsolidate)

CONFIGURATION

All execution methods and parameter defintions are controlled by a single configuration object

Get a config object with defualt parameters

set = settings()

INPUTS, OUTPUTS and FORMATS

set$files$file = <YOUR_INPUT_FILE>
set$files$odir = <YOUR_OUTPUT_DIRECTORY>
set$files$format = "fe"

Note: format "fe" is for the Agilent feature extraction file format. Simple "bed" format also supported see walkthrough

EXECUTION

Run all desired normalisation steps:

for (x in 1:length(set$norm)) {
      if (set$norm[x] == 1) {
        result <- try( do.call( names(set$norm[x]), list(set) ) )
      }
}

Run all desired CNV detection algorithms:

for (x in 1:length(set$algorithms)) {
      if (set$algorithms[x] == 1) {
        result <- try( do.call( names(set$algorithms[x]), list(set) ) )
      }
}

Run all desired combination, algorithm weighting, post processing and annotation steps:

for (x in 1:length(set$combine)) {
      if (set$combine[x] == 1) {
        result <- try( do.call( names(set$combine[x]), list(set) ) )
      }
}

OUTPUT

The final CNV calls are contained in a file with "_FinalReport.txt" extension which will be created in the ouput directory.

EXAMPLE WALKTHROUGH

Make temp directory, cd and start R:

mytmpdir=`mktemp -d 2>/dev/null || mktemp -d -t 'mytmpdir'` && cd $mytmpdir
R

Run a test is R using synthetic data example

# load library
library(CNsolidate)

# generate some test data
outputdir = getwd()
inputname = paste(outputdir, "/testData.txt", sep="")
testchr = 1
testlength = 20000
testmean = 0.001
testsd = 0.2
testproportion = 0.2
testData = syn.genome(data.frame(testchr,testlength,testmean,testsd,testproportion))

# write test data to file note: last column is a weight used for each probe (set all to 1 for this example)
write.table(data.frame(testData$data,1), file=inputname, sep="\t", row.names=F, col.names=F, quote=F)

# set default parameters
set = settings()
set$files$file = inputname
set$files$odir = outputdir
set$files$format = "bed"

# run CNV callers
for (x in 1:length(set$algorithms)) {
      if (set$algorithms[x] == 1) {
        result <- try( do.call( names(set$algorithms[x]), list(set) ) )
      }
}

# combine CNV callers
for (x in 1:length(set$combine)) {
      if (set$combine[x] == 1) {
        result <- try( do.call( names(set$combine[x]), list(set) ) )
      }
}

# read final CNV call list
finalreport = paste(substr(set$files$file, 1, nchar(set$files$file)-4), "_FinalReport.txt", sep="")
cnv_calls = read.table(finalreport)
colnames(cnv_calls) = c("chr", "start", "stop", "meanl2r", "probes", "start_index", "stop_index", "algorithms", "wscore", "adj_wscore", "p_value")

# make a crude plot
plot(testData$data[,4], pch=20, ylim=c(-3,3))
abline(v=cnv_calls$start_index, col="green")
abline(v=cnv_calls$stop_index, col="red")

For the interested

using the "testData$rep" variable (the locations of the inserted synthetic CNVs) one could compare the synthetic CNV locations to those detected by CNsolidate i.e. in the "cnv_calls" variable.