limma (version 3.28.14)

dupcor: Correlation Between Duplicates

Description

Estimate the correlation between duplicate spots (regularly spaced replicate spots on the same array) or between technical replicates from a series of arrays.

Usage

duplicateCorrelation(object, design=NULL, ndups=2, spacing=1, block=NULL, trim=0.15, weights=NULL)

Arguments

object

a numeric matrix of expression values, or any data object from which as.matrix will extract a suitable matrix such as an MAList, marrayNorm or ExpressionSet object. If object is an MAList object then the arguments design, ndups, spacing and weights will be extracted from it if available and do not have to be specified as arguments. Specifying these arguments explicitly will over-rule any components found in the data object.

design

the design matrix of the microarray experiment, with rows corresponding to arrays and columns to comparisons to be estimated. The number of rows must match the number of columns of object. Defaults to the unit vector meaning that the arrays are treated as replicates.

ndups

a positive integer giving the number of times each gene is printed on an array. nrow(object) must be divisible by ndups. Will be ignored if block is specified.

spacing

the spacing between the rows of object corresponding to duplicate spots, spacing=1 for consecutive spots

block

vector or factor specifying a blocking variable

trim

the fraction of observations to be trimmed from each end of tanh(all.correlations) when computing the trimmed mean.

weights

an optional numeric matrix of the same dimension as object containing weights for each spot. If smaller than object then it will be filled out the same size.

Value

A list with components

consensus.correlation

the average estimated inter-duplicate correlation. The average is the trimmed mean of the individual correlations on the atanh-transformed scale.

cor

same as consensus.correlation, for compatibility with earlier versions of the software

atanh.correlations

numeric vector of length nrow(object)/ndups giving the individual genewise atanh-transformed correlations.

Details

When block=NULL, this function estimates the correlation between duplicate spots (regularly spaced within-array replicate spots). If block is not null, this function estimates the correlation between repeated observations on the blocking variable. Typically the blocks are biological replicates and the repeated observations are technical replicates. In either case, the correlation is estimated by fitting a mixed linear model by REML individually for each gene. The function also returns a consensus correlation, which is a robust average of the individual correlations, which can be used as input for functions lmFit or gls.series.

At this time it is not possible to estimate correlations between duplicate spots and between technical replicates simultaneously. If block is not null, then the function will set ndups=1, which is equivalent to ignoring duplicate spots.

For this function to return statistically useful results, there must be at least two more arrays than the number of coefficients to be estimated, i.e., two more than the column rank of design.

The function may take long time to execute as it fits a mixed linear model for each gene for an iterative algorithm. It is not uncommon for the function to return a small number of warning messages that correlation estimates cannot be computed for some individual genes. This is not a serious concern providing that there are only a few such warnings and the total number of genes is large. The consensus estimator computed by this function will not be materially affected by a small number of genes.

References

Smyth, G. K., Michaud, J., and Scott, H. (2005). The use of within-array replicate spots for assessing differential expression in microarray experiments. Bioinformatics 21(9), 2067-2075. [http://bioinformatics.oxfordjournals.org/content/21/9/2067] [Preprint with corrections: http://www.statsci.org/smyth/pubs/dupcor.pdf]

See Also

These functions use mixedModel2Fit from the statmod package.

An overview of linear model functions in limma is given by 06.LinearModels.

Examples

# Simulate gene expression data for 100 probes and 6 microarrays
# Microarray are in two groups
# First two probes are more highly expressed in second group
# Std deviations vary between genes with prior df=4
sd <- 0.3*sqrt(4/rchisq(100,df=4))
y <- matrix(rnorm(100*6,sd=sd),100,6)
rownames(y) <- paste("Gene",1:100)
y[1:2,4:6] <- y[1:2,4:6] + 2
design <- cbind(Grp1=1,Grp2vs1=c(0,0,0,1,1,1))
options(digits=3)

# Fit with correlated arrays
# Suppose each pair of arrays is a block
block <- c(1,1,2,2,3,3)
dupcor <- duplicateCorrelation(y,design,block=block)
dupcor$consensus.correlation
fit1 <- lmFit(y,design,block=block,correlation=dupcor$consensus)
fit1 <- eBayes(fit1)
topTable(fit1,coef=2)

# Fit with duplicate probes
# Suppose two side-by-side duplicates of each gene
rownames(y) <- paste("Gene",rep(1:50,each=2))
dupcor <- duplicateCorrelation(y,design,ndups=2)
dupcor$consensus.correlation
fit2 <- lmFit(y,design,ndups=2,correlation=dupcor$consensus)
dim(fit2)
fit2 <- eBayes(fit2)
topTable(fit2,coef=2)