Skip to main content
. 2020 Nov 2;9(11):giaa118. doi: 10.1093/gigascience/giaa118

Table 2:

Performance of CopyDetective (raw, i.e., without optional final filtration, and filter, i.e., with default filtration threshold of 10.76) in comparison to 5 established approaches: ExomeCNV, VarScan2, ExomeDepth, Control-FREEC (WR and KS), and CNV-seq.

Tool and configuration Data set TP calls (+ false type) FP calls CNVs Sens PPV F1
Found Missed Detectable
ExomeCNV 1 1,378 (+2,204) 215,865 49 6 55 0.89 0.01 0.01
2 280 (+508) 13,213 32 15 47 0.68 0.02 0.04
3 1,017 (+1,904) 66,686 40 8 48 0.83 0.02 0.03
4 94 (+1) 2,064 24 64 88 0.27 0.04 0.08
VarScan2 1 119 (+126) 11,736 27 22 49 0.55 0.01 0.02
2 106 (+185) 2,758 26 11 37 0.70 0.04 0.07
3 65 (+16) 374 21 14 35 0.60 0.15 0.24
4 30 (+0) 54 23 65 88 0.26 0.36 0.30
ExomeDepth 1 163 (+50) 8,074 13 36 49 0.27 0.02 0.04
2 275 (+162) 2,042 25 12 37 0.68 0.12 0.20
3 175 (+33) 2,047 20 15 35 0.57 0.08 0.14
4 909 (+0) 375 32 56 88 0.36 0.71 0.48
ControlFREEC
 WR 1 7 (+6) 1,568 5 50 55 0.09 <0.01 0.01
2 6 (+3) 278 3 44 47 0.06 0.02 0.03
3 7 (+9) 654 6 42 48 0.13 0.01 0.02
4 5 (+2) 231 5 83 88 0.06 0.02 0.03
 KS 1 24 (+38) 7,261 12 43 55 0.22 <0.01 0.01
2 16 (+11) 1,124 10 38 48 0.21 0.01 0.03
3 16 (+11) 1,124 10 38 48 0.21 0.01 0.03
4 32 (+4) 224 17 71 88 0.19 0.09 0.13
CNV-seq 1 25,690 (+27,757) 1,723,974 21 34 55 0.38 0.01 0.03
2 3,016 (+1,885) 94,461 21 26 47 0.45 0.03 0.06
3 6,628 (+4,518) 316,311 19 29 48 0.40 0.02 0.04
4 786 (+1,125) 28,863 15 73 88 0.17 0.03 0.05
CopyDetective
 Raw 1 33 (+22) 729 18 1 19 0.95 0.04 0.08
2 63* (+43) 176 34 3 37 0.92 0.26 0.41
3 67 (+21) 212 40 1 41 0.98 0.24 0.39
4 23 (+23) 399 19 0 19 1.00 0.05 0.10
 Filter 10.76 1 25 (+15) 180 18 1 19 0.95 0.12 0.22
2 50 (+31) 63 34 3 37 0.92 0.44 0.60
3 60 (+18) 10 40 1 41 0.98 0.86 0.90
4 22 (+16) 63 19 0 19 1.00 0.26 0.41

The table reports true-positive (TP) calls (in parentheses: reporting the number of additional true-positive calls if CNV type is not evaluated), false-positive (FP) calls, found, missed, and detectable CNVs, sensitivity (sens; only evaluating true-positive calls with correct CNV type), positive predictive value (PPV; only evaluating TP calls with correct CNV type), and the F1 score. * Sixty-four detected CNVs are overlapping true CNVs. However, as 1 called CNV is clearly shorter than the validated one and characterized by a remarkably low quality value, we assume that this overlap is occurring by just coincidence. Therefore, it is counted as “missed."