Fig. 3. Performance comparision of xTea with MELT, Mobster, and TraFiC-mem on short reads.

a xTea and MELT show higher F1 scores on L1, Alu, and SVA than Mobster. Compared to MELT, xTea has a higher F1 score on L1 and Alu insertions and similar performance on SVA insertions. In particular, xTea has much better performance in calling L1 insertions on all different read depths. b The relationship of the 17 members of the pedigree. c Number of overlapping, MELT-specific, and xTea-specific Alu, L1, and SVA insertions in the 11 children. d The defined genotype inconsistency, where ‘F’, ‘M’, and ‘C’ indicate the genotype of father, mother, and child, respectively, and ‘0’, ‘1’, and ‘2’ represents reference homozygous, heterozygous, and homozygous alternate, respectively. e Insertions overlapping between xTea and MELT show similar genotype consistency. For nonoverlapping ones, xTea performs much better on Alu and L1 than MELT. For the boxplots, the box demarcations represent the 25, 50, and 75th percentiles, and the whiskers extend from the box to the largest and smallest data points at most 1.5 times the interquartile range away from the median. f Overlaps between xTea and TraFic-mem on somatic canonical L1 insertions and L1 transductions for 15 colon samples. g Manual inspection of all the L1 insertions detected by both xTea and TraFiC-mem through IGV screenshot. For TraFiC-mem, 28 canonical L1 insertion, and 7 transductions are checked as false positive; for xTea, 36 canonical L1, and 24 transductions are ‘validated’ as false positives.