Skip to main content

Table 1 Prediction performance evaluation using simulated structural variants

From: VALOR2: characterization of large-scale structural variants using linked-reads

Variant

Tool

# Sim.

# Pred.

TP

FP

FN

Pr.

Rec.

F1

Duplications (direct)

VALOR2

111

103

89

14

22

0.86

0.80

0.83

Duplications (inverted)

VALOR2

49

51

43

8

6

0.84

0.88

0.86

Inversions

VALOR2

90

65

54

11

36

0.83

0.60

0.70

 

VALOR1

90

63

47

13

43

0.78

0.52

0.63

 

LUMPY/smoove

90

35

27

7

63

0.79

0.30

0.44

 

DELLY

90

358

39

293

51

0.12

0.43

0.18

 

TARDIS

90

43

34

1

56

0.97

0.38

0.54

 

Sniffles

90

787

72

603

18

0.11

0.80

0.19

 

Long Ranger

90

75

54

20

36

0.73

0.60

0.66

 

Long Ranger VALOR2

90

102

70

31

20

0.69

0.78

0.73

 

Long Ranger ∩ VALOR2

90

38

38

0

52

1.00

0.42

0.59

Deletions

VALOR2

85

81

74

7

11

0.91

0.87

0.89

 

LUMPY/smoove

85

292

66

226

19

0.23

0.78

0.35

 

DELLY

85

496

72

424

13

0.15

0.85

0.25

 

TARDIS

85

152

70

82

15

0.46

0.82

0.59

 

Sniffles

85

467

72

395

13

0.15

0.85

0.26

 

Long Ranger

85

262

79

175

6

0.31

0.93

0.47

 

Long Ranger VALOR2

85

270

163

185

3

0.47

0.98

0.63

 

Long Ranger ∩ VALOR2

85

84

79

5

6

0.94

0.93

0.93

Translocations

VALOR2

38

27

27

0

11

1.00

0.71

0.83

 

LUMPY/smoove

38

4

2

2

36

0.50

0.05

0.10

 

DELLY

38

116

30

86

8

0.26

0.79

0.39

 

Long Ranger

38

29

26

3

12

0.90

0.68

0.78

 

Long Ranger VALOR2

38

38

53

3

3

0.95

0.95

0.95

 

Long Ranger ∩ VALOR2

38

18

18

0

20

1.00

0.47

0.64

  1. We evaluate the prediction performance of only large SVs (> 80 kbp for inversions, > 40 kbp for duplications, > 100 kbp for deletions, and > 100 kbp for translocations). Note that VALOR1, LUMPY, DELLY, Sniffles, and Long Ranger are not able to call interspersed duplications, and TARDIS can call duplications < 10 kb, which is smaller than the variants shown in this table. Precision is calculated as \(\frac {\text {TP}}{\text {TP+FP}}\), and recall is defined as \(\frac {\text {TP}}{\text {TP+FN}}\), where TP is the true positive, FP is the false positive, FN is the false negative, Pr. is the precision, and Rec is the recall. F1-score (shown as F1) is calculated as \(2\times \frac {\text {precision}\times \text {recall}}{\text {precision + recall}}\). SV calls predicted by both Long Ranger and VALOR2 (> 50% reciprocal overlap) are merged into a single call. Best values are highlighted with boldface font