Skip to main content

Table 6 FDRs with and without controlling batch covariates

From: UMI-count modeling and differential expression analysis for single-cell RNA sequencing

 

No filtering

 

TPM ≥ 50

 

Method

FDR

False (n)

DE (n)

FDR

False (n)

DE (n)

Monocle2

0.279

37.7

132.4

0.202

24.3

118.65

Monocle2_plateCov

0.100

10.25

101.3

0.054

5.3

96.2

MAST

0.264

28.75

74.65

0.264

28.75

74.65

MAST_plateCov

0.268

28.65

72.85

0.268

28.65

72.85

ROTS

0.258

36.2

132.1

0.206

26.75

122.1

NBID

0.326

50.35

146.85

0.183

22.9

119.05

NBID_scran

0.328

50.45

146.95

0.200

25.3

121.45

NBID_plateCov

0.124

13.5

108.1

0.048

4.75

99.05

NBID_scran_plateCov

0.124

13.45

108

0.048

4.75

98.9

  1. Simulation based on data: CEL-Seq2 (both batch A and batch B) from Ziegenhain et al. [12]. Sample size was 60 (30 cells in each group). In total, Replicate A had 30 cells and Replicate B had 35 cells after QC. Group 1 had 9 cells from Replicate A and 21 cells from Replicate B. Group 2 had 18 cells from Replicate A and 12 cells from Replicate B. Method names with plateCov indicate adjusting the batch covariates. NBID_scran and NBID_scran_plateCov used the size factor computed by scran as the offset instead of the total UMI counts
  2. Bold values indicate FDR > 0.05
  3. Bold and underlined values indicate FDR > 0.1