---

Functional normalization

Motivation

Our workflow uses the functional normalization approach¹⁶, which exploits internal control probes designed to detect technical variations without assaying biological differences.

It has been shown to perform favourably when compared to other approaches¹⁷. Using the internal control probes avoids the problems associated with global normalization methods, where biological variation can be mistaken for a technical effect and removed. This is especially important in studies where groups are expected to have differential methylation signatures, such as multiple tissue studies¹⁸.

Conversations on the best approaches for normalization in DNAm data pipelines are ongoing¹⁹.

---

Principal Components

The default of selecting only two principal components is often too low for this type of data. Often you will see a drop-off in proportion of variance explained after a certain number of principal components, and this can indicate an efficient selection.

var_explained %>% ggplot(aes(x=PC, y=var_explained)) +
  geom_line() +
  geom_point(color='grey5', fill='#6DACBC', shape=21, size=3) + 
  scale_x_continuous(breaks=1:ncol(pca$x)) +
  xlab("Principal Component") + 
  ylab("Proportion of variance explained") +
  theme_bw()

---

Running Normalization

In order to run normalization the annotation of the RGset must be updated for EPIC arrays.

RGset@annotation <- c(array = "IlluminaHumanMethylationEPIC", annotation = "ilm10b4.hg19")

By default, functional normalization returns normalized copy number data making the returned GenomicRatioSet twice the size necessary when only beta-values or M-values are required. Therefore, we developed preprocessFunnorm.DNAmArray(), which adds an option not to return these to the preprocessFunnorm() function from minfi.

RGset <- preprocessFunnorm.DNAmArray(
  rgSet = RGset,
  nPCs = 4,
  keepCN = FALSE,
  verbose = TRUE
)

You can now extract the normalized beta values from this object. It can also be advisable to round the beta values to 4 significant figures, which both reduces the data size and provides a realistic measure of the available precision.

betas <- getBeta(RGset, type="Illumina")
betas <- round(betas, 4)

If required, these can be transformed to M-values using minfi’s getM() which applies a logit2-transformation to the beta-values.

---