Evaluation of normalization methods for cDNA microarray data by k-NN classification

BMC Bioinformatics

Table 1 Single-bias-removal location normalization techniques used in this study. These strategies remove spatial- or intensity effect in a single step. The abbreviations are as follows, (for a given microarray), M_l: location-normalized log ratio; median(M): median value of non-normalized log ratios; lowess(rloc_i, cloc_i): lowess curve fitted as a function of the row location (rloc_i) and column location (cloc_i) of spots in PT group i; median(M_w): median value of non-normalized log ratios within the window size determined by w; lowess(A): lowess curve fitted to an MA plot of spots on a microarray; lowess(A_i): lowess curve fitted to an MA plot of spots in PT group i; spline(A_iset): spline curve fitted to an MA plot of spots in the invariant set, iset; R_l: location-normalized R value; qspline(G_i): qspline smoothing using geometric mean of the G channels of all arrays as a target array; G_l: location-normalized G value; qspline(R_t): qspline smoothing using geometric mean of the R channels of all arrays as a target array.

Name *	Description: Effect/Level	Bioconductor R package/function(parameters)
N ONRM	No normalization M_l= M	marray/maNorm(norm="none")
GMEDIAN	Global M_l= M - median(M)	marray/maNorm (norm="median", subset = T)
SL LOESS	Spatial/local lowess M_l= M - loess(rloc_i, cloc_i)	marray/maNormMain (f.loc = list(maNorm2D(g="maPrintTip", subset = T, span = 0.4))
SL FILTERW3	Spatial/Local median filter M_l= M - median(M_w), W = 3 × 3	tRMA/SpatiallyNormalise** (M, width = 3, height = 3)
SL FILTERW7	Spatial/Local median filter M_l= M - median(M_w), W = 7 × 7	tRMA/SpatiallyNormalise** (M, width = 7, height = 7)
IG LOESS	Intensity/Global lowess M_l= M - loess(A)	marray/maNorm (norm="loess", subset = TRUE, span = 0.4)
IL LOESS	Intensity/Local lowess M_l= M - loess(A_i)	marray/maNorm (norm="printTipLoess", subset = T, span = 0.4)
IST SPLINE	Intensity/Global spline M_l= M - spline(A_iset)	affy/normalize.invariantset**(prd.td = c(0.003, 0.007))
QSPLINEG	Intensity/Global qspline R_l= R - qspline(G_t), G_l= G - qspline(G_t), M_l= log(R_l/ G_l)	affy/R_l← normalize.qspline(R, 2^rowMeans(log2(G), na.rm = T), na.rm = T, default) G_l← normalize.qspline(G, 2^rowMeans(log2(G), na.rm = T), na.rm = T, default)
QS PLINER	Intensity/Global qspline R_l= R - qspline(R_t), G_l= G - qspline(R_t), M_l= log(R_l/ G_l)	affy/ R_l← normalize.qspline(R, 2^rowMeans(log2(R), na.rm = T), na.rm = T, default) G_l← normalize.qspline(G, 2^rowMeans(log2(R), na.rm = T), na.rm = T, default)

* We adopted the terminology given in the table to avoid confusion within this work. Elsewhere, these methods are known as: GMEDIAN, global or median [4]; SL LOESS, 2D spatial [12]; SL FILTERW3, spatial normalization using median filter of the block size 3 × 3 [17]; SL FILTERW7, spatial normalization using median filter of the block size 7 × 7 [17]; IG LOESS, global loess [4, 26]; IL LOESS, print-tip loess [4]; IST SPLINE, invariant set normalization [38]; QS PLINER, qspline using geometric mean of the R channels of all arrays as the target array [13]; Q SPLINEG, qspline using geometric mean of the G channels of all arrays as the target array [13].
** The SpatiallyNormalise function in the tRMA package was modified to remove scale normalization. The normalize.invariantset function in Affy package was modified so that the function could be applied on cDNA microarray data.
*default* The default parameters for Q SPLINEG and QS PLINER are (fit.iters = 5, min.offset = 5, spline.method="natural", smooth = T, spar = 0, p.min = 0, p.max = 1.0, incl.ends = T, converge = F)

ISSN: 1471-2105