Table 1 Normalisation methods tested. The table includes a brief description of each method, the relevant R package and reference for further information

Quantile normalisation

The distributions of probe intensities for different samples are made identical. Often used in microarray analysis.

lumi

[33]

Stratified quantile normalisation

Probes are stratified by genomic region then quantile normalised with sex chromosomes normalised separately when male and female samples are present. No background correction, zeros removed by outlier function. Not recommended for cancer-normal comparisons or other groups with global differences.

minfi

[15]

Beta-mixture quantile dilation (BMIQ)

Adjusts type II probes to type I distribution. Recommended for all datasets.

ChAMP

[27]

Subset-quantile within array normalisation (SWAN)

A quantile distribution is created using a subset of probes, with subsetting based on the number of CpGs in the probe body. Separate subsets are created for type I and II probes. The remaining probes are then adjusted to the subsets.

minfi

[34]

Functional normalisation (FunNorm)

Uses control probes to remove unwanted technical variation. Also diminishes batch effects in some datasets. Suitable for use in cancer-normal studies or where global methylation changes occur.

minfi

[29]

Dasen

Background adjustment and between array normalisation are performed separately on type I and II probes.

wateRmelon

[20]

Noob

Uses type I probe design to measure non-specific fluorescence in the opposite colour channel.

minfi

[35]

Remove unwanted variation (RUV)

Previously used with microarray data to normalise via negative control genes. Requires distinct groups such as cancer-normal to normalise on.

RUVnormalize

[36]

Batch correction: ComBat

Adjusts for known or unknown batches using an empirical Bayesian framework.

sva

[19]