Table 1.

Comparison of benchmarked deconvolution methods. Sequencing-based methods have three common criteria in the informative region selection: number of CpG, region size and read coverage. Some criteria were altered to be more suitable for the dataset we used in our analyses. For array-based methods, we specifically designated CpG sites based on methylation variance or ctDMRs. Furthermore, we described differences in cell-type composition estimation step based on three criteria again: reference-requirements, number of detectable subpopulations and estimation scope

Method	Main data type	Informative region selection			Cell-type composition estimation
		# CpG	Region size (bp)	Coverage	Class	# Components	Estimation Scope
BED	RRBS	>4	NA	>20	ref-based	2	local
ClubCpG	WGBS	>4	100	>20	ref-based	2 or more	global
csmFinder + coMethy	WGBS	>4	NA	>10	ref-free	2 or more	global
DXM	Any kinds of BS-seq	Promoter and CpG island regions		>4	ref-free	2 or more	global
MethylPurify	WGBS	>10	300/200	>10	ref-free	2	local
Prism	RRBS	>4	NA	>20	ref-free	2 or more	local
Houseman	Methylation microarrays	CpGs overlapping with ctDMRs			ref-based	2 or more	global
MeDeCom	Methylation microarrays	CpGs showing high methylation variance across pseudo-bulks			ref-free	2 or more	global

The original setup in [44] is 2 for minimum number of CpG and 10 for minimum read coverage. The original setup in [43] is 300 bp for region size. In tumor-normal pseudo-bulk analysis, we changed region size parameter value to 200 bp. BED, csmFinder and Prism do not require specific region size.