Table 2. Summary of the real data sets used in this work.

Column descriptions (in order of appearance) are as follows: (1) data set names corresponding to three published data sets and the new in-house data set; (2) type of cells in each data set; (3) number of cells considered for each data set, which varied from tens to hundreds of cells; (4) number of clusters, as reported in the respective published papers, NA (not available) for our in-house data set; (5) genomic functional regions considered for each data set, which were the same as in the original papers when applicable, CGI stands for CpG Islands, TFBS stands for Transcription Factor Binding Sites; (6) missing data proportion for each data set for the 10,000 loci filtered input and varying from 0.69 to 0.89; (7) number of loci for the largest input data sets obtained by including all regions with methylation IQR ≥ 0.01; these varied from one-quarter million to 1 million CpG sites; (8) missing data proportion for the largest input data sets, which varied from 0.54 to 0.98.

Data set	Cell type	# cells	# clusters	Regions	Miss 10K	Nloci IQR ≥ .01	Miss IQR ≥ .01
Smallwood2014 [16]	mouse embryonic stem cells	32	2	CGI	0.69	786 620	0.54
Hou2016 [11]	human hepatocellular carcinomas	25	2	CGI	0.87	255 136	0.90
Farlik2016 [12]	human hematopoietic cells	122	6	TFBS	0.89	512 153	0.98
InHouse	human xenografted cancer cells (3 patients)	558	NA	CGI	0.82	1 019 956	0.79