Table 1 Performance of HIRE and other competing methods in simulation studies

From: Detection of cell-type-specific risk-CpG sites in epigenome-wide association studies

 

Cell type number

Sample size

 

HIRE

SVA

Unadjusted

RefFreeEWAS

EWASHer

ReFACTor

True null

K = 3

n = 180

FPR

0

0.35%

21.72%

0.12%

0

0

True null

K = 3

n = 300

FPR

0.00%

3.13%

53.40%

0.12%

0.00%

0

True null

K = 3

n = 600

FPR

0

22.06%

77.2%

0.1%

0.00%

0

True null

K = 5

n = 180

FPR

0.05%

0.3%

18.5%

0.13%

0

0

True null

K = 5

n = 300

FPR

0.01%

1.54%

36.89%

0.08%

0

0

True null

K = 5

n = 600

FPR

0

8.09%

55.66%

0.12%

0

0

True null

K = 7

n = 180

FPR

0.00%

0.04%

5.13%

0.11%

0

0

True null

K = 7

n = 300

FPR

0.00%

0.22%

22.84%

0.12%

0

0

True null

K = 7

n = 600

FPR

0

6.80%

48.12%

0.11%

0

0

True alternative

K = 3

n = 180

FPR

0.01%

1.24%

32.20%

0.11%

0

0

True alternative

K = 3

n = 180

TPR

98.67%

87.33%

60.67%

79.33%

21.33%

60.67%

True alternative

K = 3

n = 300

FPR

0.00%

7.60%

65.03%

0.09%

0

2.21%

True alternative

K = 3

n = 300

TPR

96.67%

86%

79.33%

63.33%

24.67%

56.67%

True alternative

K = 3

n = 600

FPR

0.03%

16.91%

73.55%

0.11%

0.00%

1.34%

True alternative

K = 3

n = 600

TPR

100%

84%

88.67%

90%

35.33%

45.33%

True alternative

K = 5

n = 180

FPR

0.00%

2.70%

20.06%

0.14%

0

0

True alternative

K = 5

n = 180

TPR

66%

80.8%

35.2%

71.2%

6.4%

41.6%

True alternative

K = 5

n = 300

FPR

0.01%

1.76%

36.11%

0.17%

0

0.01%

True alternative

K = 5

n = 300

TPR

89.6%

86.4%

69.2%

76.4%

11.2%

53.6%

True alternative

K = 5

n = 600

FPR

0.01%

15.31%

56.25%

0.09%

0

0.11%

True alternative

K = 5

n = 600

TPR

98.4%

82.8%

62%

84.4%

18%

62.8%

True alternative

K = 7

n = 180

FPR

0

0.37%

9.30%

0.11%

0

0

True alternative

K = 7

n = 180

TPR

43%

58.67%

35%

54.67%

5%

26%

True alternative

K = 7

n = 300

FPR

0.00%

1.08%

26.67%

0.11%

0

0

True alternative

K = 7

n = 300

TPR

63.33%

73%

45%

76.33%

5%

35.67%

True alternative

K = 7

n = 600

FPR

0.04%

25.80%

56.76%

0.10%

0

0.00%

True alternative

K = 7

n = 600

TPR

82.67%

83%

66%

74.33%

5%

51.67%

  1. Performance of HIRE and other competing methods in simulation studies in detecting risk-CpG sites at the aggregate level. For the true null cases in which no CpG site is at risk, the average of the false positive rates (FPRs) based on five replicates is reported. For the true alternative cases, the averages of the FPRs and the true positive rates (TPRs) based on five replicates are reported. The number of CpG sites at risk is 30, 50, and 60 for the cell type number K = 3, 5, and 7, respectively. HIRE calls a CpG site as significant at the aggregate level if it is at risk in at least one cell type. We used Bonferroni correction for each method to control the family-wise error rate (FWER) below α = 0.01. Because HIRE can provide the p-values of CpG sites for all cell types and phenotypes, the p-value threshold for significance is α/(mKq), where m = 10,000 is the number of CpG sites and q = 2 is the phenotype number. For the other five methods, the p-value threshold is set to α/m. Notice that “0” represents exact zero, and “0.00%” indicates a very small positive number that is rounded down to zero using four decimal places
Back to article page