Harman

The removal of batch effects from datasets using a PCA and constrained optimisation based technique.

Harman is a PCA and constrained optimisation based technique that maximises the removal of batch effects from datasets, with the constraint that the probability of overcorrection (i.e. removing genuine biological signal along with batch noise) is kept to a fraction which is set by the end-user (Oytam et al, 2015).

The Harman approach can be generalised to any high dimensional dataset where an experimental factor of interest to be kept is declared and another (usually technical) factor is declared to be removed, with the constraint that the factor to be kept and removed are not completely confounded.

To compile the package on a command line:
Windows: Rcmd.exe INSTALL --preclean --no-multiarch --with-keep.source Harman
Linux: R CMD INSTALL --preclean --build Harman -l ~/R/library

To set up help files:
roxygen2::roxygenize(package.dir = "/path/to/harman")

N.B.

• The roxygenize() function should create the NAMESPACE file also.

To use harman:

install.packages("Harman")
library("Harman")
data(OLF)
expt <- olf.info$Treatment
batch <- olf.info$Batch
olf.harman <- harman(olf.data, expt, batch)
olf.data.corrected <- reconstructData(olf.harman)

Further documentation and the Matlab version is available here: http://www.bioinformatics.csiro.au/harman/