evaluateCorrection: Quality assessment of a batch correction

In rwehrens/BatchCorrMetabolomics: Complementary package to "Improved Batch Correction in Untargeted MS-Based Metabolomics"

Description

Two criteria are employed to assess the quality of a batch correction. The first is based on a PCA decomposition; the other on the variation between (biological) replicates. This function is a wrapper for the two underlying functions - use these if you want to have access to more specific arguments.

Usage

evaluateCorrection(X, Y, what = c("duplo", "PCA"), ...)

Arguments

X	Data matrix, with samples in the rows and variables in the columns.
Y	A vector of batch labels.
what	Name of the quality criterion to be employed.
...	Further arguments to graphical functions.

Value

A number indicating the relative success of the batch correction. For the PCA criterion, the number is the average Bhattacharya distance between the batches - smaller is better. For the duplo criterion, the number means the relative fraction of variation accounted for by the sample (in our example the genotype): bigger is better.

Author(s)

Ron Wehrens

References

" @ArticleWehrens2016, author = Ron Wehrens and Jos.~A.~Hageman and Fred~van~Eeuwijk and Rik~Kooke and P\'adraic~J.~Flood and Erik Wijnker and Joost~J.B.~Keurentjes and Arjen~Lommen and Henri\"ette~D.L.M.~van~Eekelen and Robert~D.~Hall and Roland~Mumm and Ric~C.H.~de~Vos, title = Improved batch correction in untargeted MS-based metabolomics, journal = Metabolomics, year = 2016, volume = 12, DOI = 10.1007/s11306-016-1015-8, pages = 1–12

"

See Also

evaluateDuplos, evaluatePCA

Examples

data(BC)
set.1.lod <- min(set.1[!is.na(set.1)])

## Figure 2 from the paper
par(mfrow = c(1,2))
## uncorrected data
huhnPCA <- evaluateCorrection(set.1, set.1.Y, what = "PCA",
                              plot = TRUE, legend.loc = "bottomright")
title(main = paste("Interbatch distance:", round(huhnPCA, 3)))

## do correction
set.1.corrected.Q <-
  apply(set.1, 2, doBC, ref.idx = which(set.1.Y$SCode == "ref"),
        batch.idx = set.1.Y$Batch, minBsamp = 4,
        seq.idx = set.1.Y$SeqNr, method = "lm",
        imputeVal = set.1.lod)
huhnPCA.A <- evaluateCorrection(set.1.corrected.Q, set.1.Y, what = "PCA",
                                plot = TRUE, legend.loc = "bottomright")
title(main = paste("Q: Interbatch distance:", round(huhnPCA.A, 3)))

rwehrens/BatchCorrMetabolomics documentation built on May 28, 2019, 10:42 a.m.