An integrative probabilistic model for identification of structural variation in sequencing data

Table 2 Inversion prediction in individual NA12878

Method	Minimum number to detect 1	Minimum number to detect 2	Minimum number to detect 3
High quality alignments
Breakdancer	47 (37)	80 (221)	NA (NA)
GASV	34 (158)	76 (298)	5,028 (NA)
GASVPro-HQ	11 (20)	116 (102)	206 (346)
Hydra-HQ	61 (139)	108 (246)	284 (NA)
All alignments
GASVPro	28 (59)	394 (286)	550 (504)
GASVPro-Min2	28 (59)	160 (334)	NA (NA)
Hydra	159 (258)	NA (470)	NA (NA)

We report the results on both Pilot 2 and Pilot 1, with the Pilot 1 results in parentheses. In most cases the sensitivity of inversion detection increases with coverage with more methods correctly predicting three inversions in the higher coverage Pilot 2 data. In some cases, true inversions identified by the uniquely mapped data are lost with the addition of ambiguous alignments. These alignments result in substantially more predictions, which can cause true inversions to be eliminated in the pruning process. The benefit of our probabilistic method and inclusion of the beRD signal is evident as higher specificity is attained by GASVPro and GASVPro-HQ compared to GASV, when predicting the top inversion. NA, not applicable.

ISSN: 1474-760X