Table 1 CpG selection options available in DecompPipeline

VAR

Most highly variable across the samples

n.markers to determine the number of sites used

Yes

Yes

Yes

RANDOM

Random subset

Yes

Yes

Yes

HYBRID

Half as most highly variable, half randomly

Yes

Yes

Yes

PCA

Highest loadings on the first n.prin.comp principal components

Default n.prin.comp = 10

Yes

Yes

Yes

PCADAPT

Principal component analysis implemented in the ‘bigstatsr’ R package

Privé et al.⁷⁴

Yes

Yes

Yes

ALL

All that fulfill the quality criteria

Yes

Yes

Not recommended

RANGE

Largest dynamic range across the samples

Yes

Yes

Yes

CUSTOM

User-specified list

Yes

Yes

Yes (recommended)

ROWFSTAT

Linked to given reference profiles using the F statistics

Requires reference profiles

Yes

Yes

Yes

PHENO

Differentially methylated according to specified phenotypic groups using the ‘limma method’

Ritchie et al.⁷⁵

Yes

Yes

Yes

HOUSEMAN2012

50,000 sites determined to be cell type specific using the Houseman et al.¹¹ method and the Reinius et al.³⁶ reference dataset.

Applicable only to blood datasets

Houseman et al.¹¹, Reinius et al.³⁶

Yes

No

Not implemented

HOUSEMAN2014

According to the RefFreeEWAS method

Houseman et al.⁷⁶

Yes

Yes

Yes

JAFFE2014

600 sites listed as cell type specific in Jaffe et al.⁷⁷

Applicable only to blood datasets

Jaffe et al.⁷⁷

Yes

No

Not implemented

EDEC_STAGE0

According to Stage 0 of the ‘EDec’ approach

Requires reference profiles, Onuchic et al.²²

Yes

Yes

No