BERMUDA: a novel deep transfer learning method for single-cell RNA sequencing batch correction reveals hidden high-resolution cellular subtypes

Table 1 Datasets used for evaluation of BERMUDA

Dataset	Batch	Protocol	Number of cells	Cell types (number of cells)
2D Gaussian	Batch1	Simulation	2000	Type1 (855), Type2 (237), Type3 (373), Type4 (535)
2D Gaussian	Batch2	Simulation	2000	Type1 (379), Type2 (656), Type3 (102), Type4 (863)
Splatter	Batch1	Splatter	2000	Type1 (810), Type2 (616), Type3 (388), Type4 (186)
Splatter	Batch2	Splatter	1000	Type1 (439), Type2 (262), Type3 (201), Type4 (98)
Pancreas	Muraro	CEL-Seq2	2042	alpha (812), beta (448), gamma (101), delta (193), epsilon (3), acinar (219), ductal (245), endothelial (21)
	Baron	inDrop	8012	alpha (2326), beta (2525), gamma (255), delta (601), epsilon (18), acinar (958), ductal (1077), endothelial (252)
	Segerstolpe	SMART-seq2	2061	alpha (886), beta (270), gamma (197), delta (114), epsilon (7), acinar (185), ductal (386), endothelial (16)
PBMC	PBMC	10x Chromium	8381
PBMC	Pan T cell	10x Chromium	3555

For the consistency in the paper, we refer to each pancreas and PBMC dataset, e.g., the Muraro dataset, as a batch, and refer to multiple batches investigating similar biological systems, e.g., pancreas, as a dataset. Type1 to Type4 in the 2D Gaussian dataset and the Splatter dataset refer to the virtual cell types that were generated using simulation techniques

ISSN: 1474-760X