Package 'MDimNormn'

Multi-dimensional MA normalization for plate effect

Description

Normalize data to minimize the difference among the subgroups of the samples generated by experimental factor such as multiple plates (batch effects)
- the primary method is Multi-MA, but other fitting function, f in manuscript (e.g. loess) is available, too.
This method is based on the assumptions stated below

1. The geometric mean value of the samples in each subgroup (or plate) for a single target is ideally same as those from the other subgroups.

2. The subgroup (or plate) effects that influence those mean values for multiple observed targets are dependent on the values themselves. (intensity dependent effects)

Usage

normn_MA(mD, expGroup, represent_FUN= function(x) mean(x, na.rm= T), 
        fitting_FUN= NULL, isLog= TRUE)

Arguments

mD	a matrix of measured values in which columns are the measured molecules and rows are samples
expGroup	a vector of experimental grouping variable such as plate. The length of codeexpGroup must be same as the number of rows of mD.
represent_FUN	a function that computes representative values for each experimental group (e.g. plate). The default is mean ignoring any NA
fitting_FUN	NULL or a function that fits to data in MA-coordinates. If it is NULL as the default, 'Multi-MA' method is employed. If a function is used, two arguments of m_j and A are required, which are $\mathbf{m}_j$ coordinate in $M_d$ and $A$ coordinate, respectively.
isLog	TRUE or FALSE, if the normalization should be conducted after log-transformation. The affinity proteomics data from suspension bead arrays is recommended to be normalized using the default, isLog = TRUE.

Value

The data after normalization in a matrix

Author(s)

Mun-Gwan Hong <[email protected]>

References

Hong M-G, Lee W, Pawitan Y, Schwenk JM (201?) Multi-dimensional normalization of plate effects for multiplexed applications unpublished

Examples

data(sba)
B <- normn_MA(sba$X, sba$plate)		# Multi-MA normalization

# MA-loess normalization
B <- normn_MA(sba$X, sba$plate, fitting_FUN= function(m_j, A) loess(m_j ~ A)$fitted)

# weighted linear regression normalization
B <- normn_MA(sba$X, sba$plate, fitting_FUN= function(m_j, A) {
	beta <- lm(m_j ~ A, weights= 1/A)$coefficients
	beta[1] + beta[2] * A
})

# robust linear regression normalization
if(any(search() == "package:MASS")) {	# excutable only when MASS package was loaded.
	B <- normn_MA(sba$X, sba$plate, fitting_FUN= function(m_j, A) {
		beta <- rlm(m_j ~ A, maxit= 100)$coefficients
		beta[1] + beta[2] * A
	})
}

---

Artificially generated data alike that of Suspension bead arrays

Description

The data that has similarity to Suspension bead arrays data.

Usage

data(sba)

Format

A list that consists of "plate" which is a factor of plate number, "X" that contains measured values where columns are targets and rows are samples (or observations).

Examples

data(sba)

# plot to check difference of geometric mean of every target between plates
sba_gm <- by(sba$X, sba$plate, apply, 2, function(x) exp(mean(log(x))))	
par(mfrow= c(2, 3))
apply(combn(4, 2), 2, function(ea) {
	plot(sba_gm[[ea[1]]], sba_gm[[ea[2]]], xlab= names(sba_gm)[ea[1]], 
	     ylab= names(sba_gm)[ea[2]], log= "xy", asp= 1)
	abline(0, 1, col= "cadetblue")
})

# show first 10 observations in plate 1 and plate 2
print(sba$X[c(1:10, 97:106), 1:10])

---