scIBD: a self-supervised iterative-optimizing model for boosting the detection of heterotypic doublets in single-cell chromatin accessibility data

Table 1 Benchmark datasets

Data source	Dataset identity	Dataset type	Number of droplets	Doublet rate	Med. of sequencing depth	Med. of sparsity	Number of peaks	Number of cell types	Complexity (silhouette coefficient)	Valid read pairs	Avg. valid read pairs per droplet	Med. valid read pairs per droplet
MF [19] (GSE100033)	SimATAC-balance	Fully-synthetic	4500	0.11	1172	0.9916	140,102	8	0.66	-	-	-
MF [19] (GSE100033)	SimATAC-imbalance	Fully-synthetic	2100	0.2	1291	0.9908	140,102	8	0.56	-	-	-
HMC [6] (GSE162690)	High-loading	Real	20,663	0.29	1102	0.9941	187,075	10	0.31	175,950,050	8515	7510
HMC [6] (GSE162690)	Low-loading	Real	12,793	0.21	1004	0.9940	166,360	10	0.42	77,050,576	6022	4641
MF [19] (GSE100033)	Forebrain*	Semi-synthetic	1298	0.2	1930	0.9915	226,759	5	0.23	40,227,305	30,992	25,004
MCA [20] (GSE111586)	Bone marrow*		5244	0.2	2120	0.9917	254,413	18	-0.22	173,189,105	33,026	20,842
	Lung*		6146	0.2	1406	0.9918	171,002	22	-0.08	130,550,777	21,241	14,293
	Whole brain*		6494	0.2	2404	0.9873	189,933	21	0.01	255,248,909	39,305	21,679
	Cerebellum		2733	0.2	1465	0.9934	221,657	20	0.04	53,152,804	19,448	10,464
	Heart		9180	0.2	2029	0.9886	178,104	22	-0.30	236,918,887	25,808	15,143
	Kidney		7717	0.2	2702	0.9858	190,690	26	-0.24	215,602,830	27,938	19,557
	Prefrontal cortex		7150	0.2	4863	0.9765	206,546	22	-0.19	307,767,321	43,044	27,204
	Spleen		4824	0.2	2439	0.9845	157,363	15	-0.02	128,524,323	26,620	19,567
Islets [16] (GSE165212)	Islet1		5076	0.2	2494	0.9830	147,104	5	0.16	102,622,313	15,495	18,966
Islets [16] (GSE165212)	Islet2		5145	0.2	2778	0.9804	138,549	5	0.13	101,595,484	17,755	12,137
10xPBMC [21, 22] (10xGenomics)	PBMC		9732	0.2	3112	0.9667	93,446	17	0.25	316,811,190	32,553	32,535

Asterisk (*) symbol implies that there are two sequencing replicates of this tissue: one is used to reserve singlets; the other one is used to generate doublets. The doublet rate values of semi-synthetic datasets indicate the simulated ratio based on the number of singlets, not the actual doublet rate of the dataset. The fully-synthetic datasets were directly simulated based on the count matrices, thus the read statistics are not available

ISSN: 1474-760X