CIDR: Ultrafast and accurate clustering through imputation for single-cell RNA-seq data

Table 3 Comparison of runtime and clustering accuracy (measured by adjusted rand index) between CIDR and four other algorithms on a simulation data set with 10,000 cells

Simulation (10K)	CIDR	CIDR (L)	prcomp	t-SNE	RaceID	ZIFA
Time	44.5 min	1.5 h	3.1 h	21.8 h	>14 day	1.6 day^a
Adjusted rand index	0.99	1.00	0.99	0.00	N/A^b	0.09

CIDR is the default CIDR algorithm implementation with step function simplification, while CIDR (L) is the implementation with the non-simplified logistic function. The algorithms except ZIFA were run on an AWS ec2 r3.2xlarge instance
^aZIFA ran out of memory on the AWS ec2 r3.2xlarge instance, and its runtime was recorded from a run on an AWS ec2 r3.8xlarge instance
^bRaceID did not complete after 14 days

ISSN: 1474-760X