normalizeBySizeFactors {MAnorm2}

Normalize ChIP-seq Samples by Their Size Factors

Description

Given read counts from a set of ChIP-seq samples in a set of genomic intervals, this function normalizes the counts using size factors of the samples, and converts the normalized read counts into normalized signal intensities more of a continuous variable. The function can also be used to normalize RNA-seq samples, in which case each genomic interval refers to a gene. In fact, the normalization method implemented in this function is most suited to RNA-seq datasets. See normalize for a more robust method for normalizing ChIP-seq samples.

Usage

normalizeBySizeFactors(
  x,
  count,
  subset = NULL,
  interval.size = FALSE,
  offset = 0.5,
  convert = NULL
)

Arguments

x

A data frame containing the read count variables. Each row should represent a genomic interval or a gene. Objects of other types are coerced to a data frame.

count

A vector of either integers or characters indexing the read count variables in x to be normalized. Each of these variables represents a ChIP-seq/RNA-seq sample. Elements of count must be unique.

subset

An optional vector specifying the subset of intervals or genes to be used for estimating size factors. For ChIP-seq samples, you may want to use only the intervals occupied by all the samples to estimate their size factors (see "Examples" below). By default, all genomic intervals or genes are used.

interval.size

A numeric vector of interval sizes or a logical scalar to specify whether to use interval sizes for converting normalized read counts into normalized signal intensities (see "Details"). If set to TRUE, the function will look for the "start" and "end" variables in x, and use them to calculate interval sizes. By default, interval sizes are not used.

In cases of analyzing RNA-seq samples, interval sizes, if used, should be the corresponding gene lengths (or sums of exon lengths).

offset

The offset value used for converting normalized read counts into normalized signal intensities (see "Details"). The default value is suited to most cases. If you are analyzing RNA-seq samples and intended to use gene lengths, however, a smaller offset value (e.g., 0.01) is recommended.

convert

An optional function specifying the way that normalized read counts are converted into normalized signal intensities. It should accept a vector of inputs and return a vector of the corresponding signal intensities. If set, interval.size and offset are ignored.

Details

This function first estimates the size factor of each sample specified, which quantifies the sample's relative sequencing depth. Technically, the function applies the median ratio method to the raw read counts, which is originally devised to normalize RNA-seq samples (see "References"). Then, normalized read counts are deduced by dividing the raw counts of each sample by its size factor.

These normalized read counts are then converted into normalized signal intensities more of a continuous variable. By default, the function uses the equation log2(normCnt + offset), or log2(normCnt / intervalSize + offset) if interval sizes (or gene lengths) are provided. To be noted, while the interval sizes (either specified by users or calculated from the data frame) are considered as number of base pairs, the intervalSize variable used in the latter equation has a unit of kilo base pairs. In this case, 0.5 still serves as a generally appropriate offset for ChIP-seq samples. For RNA-seq samples, however, a smaller offset value (e.g., 0.01) should be adopted.

In most cases, simply using the former equation is recommended. You may, however, want to involve the interval sizes (or gene lengths) when the samples to be classified into the same biological condition are associated with a large variation (e.g., when they are from different individuals; see also bioCond). Besides, the goodness of fit of mean-variance curve (see also fitMeanVarCurve) could serve as one of the principles for selecting an appropriate converting equation.

The convert argument serves as an optional function for converting normalized read counts into normalized signal intensities. The function is expected to operate on the vector of normalized counts of each sample, and should return the converted signal intensities. convert is barely used, exceptions including applying a variance stabilizing transformation or shrinking potential outliers.

Value

normalizeBySizeFactors returns the provided data frame, with the read counts replaced by the corresponding normalized signal intensities. Besides, an attribute named "size.factor" is added to the data frame, recording the size factor of each specified sample.

References

Anders, S. and W. Huber, Differential expression analysis for sequence count data. Genome Biol, 2010. 11(10): p. R106.

See Also

normalize for performing an MA normalization on ChIP-seq samples; estimateSizeFactors for estimating size factors of ChIP-seq/RNA-seq samples; MAplot for creating an MA plot on normalized signal intensities of two samples; bioCond for creating an object to represent a biological condition given a set of normalized samples, and normBioCondBySizeFactors for normalizing such objects based on their size factors.

Examples

data(H3K27Ac, package = "MAnorm2")
attr(H3K27Ac, "metaInfo")

## Normalize directly the whole set of ChIP-seq samples by their size
## factors.

# Use only the genomic intervals that are occupied by all the ChIP-seq
# samples to be normalized to estimate the size factors.
norm <- normalizeBySizeFactors(H3K27Ac, 4:8,
                               subset = apply(H3K27Ac[9:13], 1, all))

# Inspect the normalization effects.
attr(norm, "size.factor")
MAplot(norm[[4]], norm[[5]], norm[[9]], norm[[10]],
       main = "GM12890_rep1 vs. GM12891_rep1")
abline(h = 0, lwd = 2, lty = 5)

## Alternatively, perform the normalization first within each cell line, and
## then normalize across cell lines. In practice, this strategy is more
## recommended than the aforementioned one.

# Normalize samples separately for each cell line.
norm <- normalizeBySizeFactors(H3K27Ac, 4)
norm <- normalizeBySizeFactors(norm, 5:6,
                               subset = apply(norm[10:11], 1, all))
norm <- normalizeBySizeFactors(norm, 7:8,
                               subset = apply(norm[12:13], 1, all))

# Construct separately a bioCond object for each cell line, and normalize
# the resulting bioConds by their size factors.
conds <- list(GM12890 = bioCond(norm[4], norm[9], name = "GM12890"),
              GM12891 = bioCond(norm[5:6], norm[10:11], name = "GM12891"),
              GM12892 = bioCond(norm[7:8], norm[12:13], name = "GM12892"))
conds <- normBioCondBySizeFactors(conds)

# Inspect the normalization effects.
attr(conds, "size.factor")
MAplot(conds[[1]], conds[[2]], main = "GM12890 vs. GM12891")
abline(h = 0, lwd = 2, lty = 5)