Scalable non-negative matrix tri-factorization

Table 1 Summary of datasets

Dataset	Database	Rows	Columns	Shape	Data type	Density	Nonzero
Fetus	GIANT [29]	25,569	25,608	rectangular	sparse	4.7%	31M
TCGA-BRCA	GDC [25]	1,222	60,483	wide	dense	100.0%	74M
E-TABM-185	ArrayExpress [28]	5,896	22,283	tall	dense	100.0%	131M
Retina	GIANT [29]	25,823	25,822	rectangular	dense	22.0%	147M
Cochlea	GIANT [29]	25,824	25,824	rectangular	dense	42.0%	280M
TCGA-Methyl	GDC [25]	10,181	485,577	wide	dense	81.4%	3841M

We manually categorized each data matrix into three shapes: tall datasets have substantially more rows than columns, wide datasets vice versa, and rectangular datasets have a comparable number of rows and columns. Density denotes the fraction of nonzero matrix elements. The number of nonzero elements in each matrix is given in the last column

ISSN: 1756-0381