calcNormFactors: Library Size Normalization

In OliverVoogd/edgeR: Empirical Analysis of Digital Gene Expression Data in R

Library Size Normalization

Description

Calculate scaling factors to convert raw library sizes into effective library sizes.

Usage

## S3 method for class 'DGEList'
calcNormFactors(object,
                method = c("TMM","TMMwsp","RLE","upperquartile","none"),
                refColumn = NULL, logratioTrim = .3, sumTrim = 0.05, doWeighting = TRUE,
                Acutoff = -1e10, p = 0.75, ...)
## S3 method for class 'SummarizedExperiment'
calcNormFactors(object,
                method = c("TMM","TMMwsp","RLE","upperquartile","none"),
                refColumn = NULL, logratioTrim = .3, sumTrim = 0.05, doWeighting = TRUE,
                Acutoff = -1e10, p = 0.75, ...)
## Default S3 method:
calcNormFactors(object, lib.size = NULL,
                method = c("TMM","TMMwsp","RLE","upperquartile","none"),
                refColumn = NULL, logratioTrim = .3, sumTrim = 0.05, doWeighting = TRUE,
                Acutoff = -1e10, p = 0.75, ...)

Arguments

object	a matrix of raw (read) counts, or a DGEList object, or a SummarizedExperiment object.
lib.size	numeric vector of library sizes corresponding to the columns of the matrix object.
method	normalization method to be used.
refColumn	column to use as reference for method="TMM". Can be a column number or a numeric vector of length nrow(object).
logratioTrim	the fraction (0 to 0.5) of observations to be trimmed from each tail of the distribution of log-ratios (M-values) before computing the mean. Used by method="TMM" for each pair of samples.
sumTrim	the fraction (0 to 0.5) of observations to be trimmed from each tail of the distribution of A-values before computing the mean. Used by method="TMM" for each pair of samples.
doWeighting	logical, whether to use (asymptotic binomial precision) weights when computing the mean M-values. Used by method="TMM" for each pair of samples.
Acutoff	minimum cutoff applied to A-values. Count pairs with lower A-values are ignored. Used by method="TMM" for each pair of samples.
p	numeric value between 0 and 1 specifying which quantile of the counts should be used by method="upperquartile".
...	other arguments are not currently used.

Details

This function computes scaling factors to convert observed library sizes into effective library sizes. The effective library sizes for use in downstream analysis are lib.size * norm.factors where lib.size contains the original library sizes and norm.factors is the vector of scaling factors computed by this function.

The TMM method implements the trimmed mean of M-values method proposed by Robinson and Oshlack (2010). By default, the M-values are weighted according to inverse variances, as computed by the delta method for logarithms of binomial random variables. If refColumn is unspecified, then the column whose count-per-million upper quartile is closest to the mean upper quartile is set as the reference library.

The TMMwsp method stands for "TMM with singleton pairing". This is a variant of TMM that is intended to perform better for data with a high proportion of zeros. In the TMM method, genes that have zero count in either library are ignored when comparing pairs of libraries. In the TMMwsp method, the positive counts from such genes are reused to increase the number of features by which the libraries are compared. The singleton positive counts are paired up between the libraries in decreasing order of size and then a slightly modified TMM method is applied to the re-ordered libraries. If refColumn is unspecified, then the column with largest sum of square-root counts is used as the reference library.

RLE is the scaling factor method proposed by Anders and Huber (2010). We call it "relative log expression", as median library is calculated from the geometric mean of all columns and the median ratio of each sample to the median library is taken as the scale factor.

The upperquartile method is the upper-quartile normalization method of Bullard et al (2010), in which the scale factors are calculated from the 75% quantile of the counts for each library, after removing genes that are zero in all libraries. The idea is generalized here to allow normalization by any quantile of the count distributions.

If method="none", then the normalization factors are set to 1.

For symmetry, normalization factors are adjusted to multiply to 1. Rows of object that have zero counts for all columns are removed before normalization factors are computed. The number of such rows does not affect the estimated normalization factors.

Value

If object is a matrix, then the output is a vector with length ncol(object) giving the library scaling factors.

If object is a DGEList or SummarizedExperiment object, then the output is a DGEList the same as input with the library scaling factors stored as object$samples$norm.factors.

If object is a SummarizedExperiment object, then it is converted to a DGEList using SE2DGEList and the DGEList method applied.

Author(s)

Mark Robinson, Gordon Smyth, Yunshun Chen.

References

Anders, S, Huber, W (2010). Differential expression analysis for sequence count data Genome Biology 11, R106.

Bullard JH, Purdom E, Hansen KD, Dudoit S. (2010) Evaluation of statistical methods for normalization and differential expression in mRNA-Seq experiments. BMC Bioinformatics 11, 94.

Robinson MD, Oshlack A (2010). A scaling normalization method for differential expression analysis of RNA-seq data. Genome Biology 11, R25.

See Also

effectiveLibSizes, SE2DGEList.

Examples

y <- matrix( rpois(1000, lambda=5), nrow=200 )
calcNormFactors(y)

# The TMMwsp and TMM methods ignore genes with largest fold-changes:
y <- cbind(1,c(1,1,1,1,1,1,1,1,1,100))
calcNormFactors(y, lib.size=c(1e6,1e6))

# calcNormFactors makes the fold-changes for the majority of genes as close to 1 as possible:
# In this example, normalizing the library sizes makes most of the CPMs equal in the two samples:
dge <- DGEList(counts=y)
cpm(dge)
dge <- calcNormFactors(dge)
cpm(dge)

OliverVoogd/edgeR documentation built on July 28, 2022, 10:13 p.m.