scDesign2: a transparent simulator that generates high-fidelity single-cell gene expression count data with gene correlations captured

Table 1 Summary of 16 simulators (including our proposed scDesign2) in six properties

Property	1	2	3	4	5	6
	protocol adaptive	gene preserved	gene cor. captured	cell num. seq. dep. flexible	transparent	comp. and sample efficient
Simulator
dyngen [77]		✗	✗		✓	✓
Lun2 [78]		✓	✗	✓	✓	✓
powsimR [75]	✓	✓	✗	✓	✓	✓
PROSST [68]		✓	✗		✓	✓
scDD [74]	✓	✗	✗		✓	✓
scDesign [35]	✓		✗	✓	✓	✓
scGAN [72]	✓	✓			✗	✗
splat simple [69]	✓	✗	✗	✗	✓	✓
splat [69]	✓	✗	✗	✗	✓	✓
kersplat [69]	✓	✗		✗	✓	✓
SPARSim [71]	✓	✓		✗	✓	✓
SPsimSeq [79]	✓	✓	✓	**	✓	✓
SymSim [70]	✓	✗	✗	✗	✓	✓
ZINB-WaVE [39]	✓			✗	✓	✓
SERGIO [76]	✓		✗ ^∗	✓	✓	✓
scDesign2	✓	✓	✓	✓	✓	✓

Property 1: protocol adaptiveness
Property 2: gene preservation
Property 3: gene correlation capture
Property 4: flexible cell number and sequencing depth choices
Property 5: transparency
Property 6: computational and sample efficiency
For each simulator and each property, a checkmark, checkcross, or cross means that the simulator satisfies, partially satisfies, or does not satisfy the property, respectively
*SERGIO requires a user-specified gene regulatory network, and it does not capture/estimate gene correlations from a real dataset
**SPsimSeq can vary cell number but not sequencing depth

ISSN: 1474-760X