normn_MA: Multi-dimensional MA normalization for plate effect
In Rundmus/BAf-R_package: Binder Array data hanlding package in facility

Description Usage Arguments Value Author(s) References Examples

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 the reference (e.g. loess) is available, too.
This method is based on the assumptions stated below

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

This function calls the normn_MA of the package MDimNormn.

normn_MA(
  baf,
  expGroupVar = batch_colname(baf, "sample"),
  represent_FUN = function(x) mean(x, na.rm = T),
  fitting_FUN = NULL,
  isLog = TRUE
)

`baf`	a `BAf-class` object
`expGroupVar`	the column name in @sinfo which contains the experimental factor to be used for grouping
`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 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`.

The BAf-class object with normalized values

Mun-Gwan Hong mun-gwan.hong@scilifelab.se

Hong M-G, Lee W, Nilsson P, Pawitan Y, & Schwenk JM (2016) Multidimensional normalization to minimize plate effects of suspension bead array data. J. Proteome Res., 15(10) pp 3473-80.

data(sba)
B <- sba[sba@sinfo$cohort != "EMPTY", ]
C <- normn_MA(B)		# normalize excluding "EMPTY"

i_rnd <- sample(1:ncol(C), 1)
sd(log(sX(B)[, i_rnd]))
sd(log(sX(C)[, i_rnd]))         # reduced variation

# MA-loess normalization
C1 <- normn_MA(B, fitting_FUN= function(m_j, A) loess(m_j ~ A)$fitted)

# On MA coordinates, weighted linear regression normalization
C2 <- normn_MA(B, fitting_FUN= function(m_j, A) {
	beta <- lm(m_j ~ A, weights= 1/A)$coefficients
	beta[1] + beta[2] * A
})

# On MA coordinates, robust linear regression normalization
if(any(search() == "package:MASS")) {	# excutable only when MASS package was loaded.
  C3 <- normn_MA(B, fitting_FUN= function(m_j, A) {
		 beta <- rlm(m_j ~ A, maxit= 100)$coefficients
		 beta[1] + beta[2] * A
	 })
}