Discovery of large genomic inversions using long range information

BMC Genomics

Table 1 Simulation results on GRCh37

Tool	No. calls	TP	FP	FN	found	precision	recall
Inversions > 40 Kbp (n = 275)
DELLY	780	461	319	19	256	59.10%	96.04%
LUMPY	174	174	0	84	191	100.00%	67.44%
GASVPro	475	9	166	266	9	1.89%	3.83%
Valor	198	191	7	54 (28)^a	221	96.46%	77.96% (87.21%^b)
Inversions > 80 Kbp (n = 167)
DELLY	589	274	315	15	152	46.52%	94.81%
LUMPY	95	95	0	61	106	100.00%	60.90%
GASVPro	404	5	399	164	3	1.24%	2.96%
Valor	131	124	7	28 (8)^a	139	94.66%	81.58% (93.94%^b)

We implanted 686 inversions to the reference genome (GRCh37) using VarSim and simulated two libraries, one pooled fosmid clone sequencing library for Valor, and one WGS data set. 275 inversions had size >40 Kbp, and 167 were >80 Kbp. ^a26 inversions (>40 Kbp) and 20 inversions (>80 Kbp) had no clone coverage. ^bwhen inversions that had no clone coverage at breakpoints are removed. TP true positive, FP false positive, FN false negative. found: number of simulated inversions that intersect (>50% reciprocal) with calls. Precision: positive predictive value, calculated as TP/(TP + FP). Recall: sensitivity, calculated as TP/(TP + FN). Note that due to diploid simulated inversions, one call may intersect with multiple implanted inversions

ISSN: 1471-2164