PeakError: compute the label error rate of peak calling models
https://travis-ci.org/tdhock/PeakError.png?branch=master
After constructing a database of labels that indicate specific samples and genomic regions with and without peaks in a ChIP-seq experiment, you can compute the label error rate of a peak calling model using this package. This is useful for
- training peak calling model parameters: choose the parameter values that minimize the number of incorrect labels.
- testing different peak calling models: compute the label error rate of two or more different peak models. The one with the lowest label error rate is the most accurate in your data.
For more details, please read our paper "Optimizing ChIP-seq peak detectors using visual labels and supervised machine learning" by Hocking et al, Bioinformatics 2017. https://www.ncbi.nlm.nih.gov/pubmed/27797775
if(!require(devtools))install.packages("devtools") devtools::install_github("tdhock/PeakError")
library(PeakError) example(PeakError)
The exampleData directory contains command line scripts for computing the PeakError for peaks and labels defined in bed files:
cd $(Rscript -e 'cat(system.file("exampleData", package="PeakError"))') Rscript PeakError_compute.R peaks.bed labels.bed > errors.tsv Rscript PeakError_summarize.R errors.tsv
On my system I got the following output:
thocking@silene:~/R$ cd $(Rscript -e 'cat(system.file("exampleData", package="PeakError"))')
thocking@silene:~/lib/R/library/PeakError/exampleData$ Rscript PeakError_compute.R peaks.bed labels.bed > errors.tsv
read 28 peaks
read 20 labels
Loading required package: PeakError
computed errors for 20 labels
thocking@silene:~/lib/R/library/PeakError/exampleData$ Rscript PeakError_summarize.R errors.tsv
noPeaks peakEnd peaks peakStart
5 5 5 5
11 / 20 incorrect labels = 55.000000% error rate
7 false positives / 15 possible = 46.666667% false positive rate
4 false negatives / 15 possible = 26.666667% false negative rate
thocking@silene:~/lib/R/library/PeakError/exampleData$