Fig. 1.

Receiver operating characteristic curves describing deletion detection in NA12878 from three scenarios. The relative accuracy of Hydra-Multi (red) was compared with both DELLY (blue and purple) and GASVPro (green) in three analyses that each compared fragment size parameters of 5 and 8 median absolute deviations (MADs) (Supplementary Methods). Each plot displays the relationship between the number of true and false positives at varying levels of minimum alignment support (4–10 read-pairs). A true positive was defined as detection of one of the 3077 non-overlapping truth set deletions where both intervals from a predicted deletion breakpoint intersected with both of the truth set deletion breakpoint intervals. In order to make a fair comparison across all tools, each predicted breakpoint was represented as two 200 bp intervals that faithfully represent the region implicated by the original SV call. A list of regions to exclude based on excessively high read-depth were used on both the truth set and putative call sets (Supplementary Methods). The three situations used to assess the three tools are as follows: (A) The 50× NA12878 dataset was subsampled to 5× and analyzed. (B) The 50× NA12878 data were analyzed. (C) The subsampled 5× NA12878 dataset was analyzed concurrently with 64 randomly selected datasets of ∼5× coverage from 1KGP. Total support was evaluated as the total number of read-pairs across all datasets analyzed. The presence of a deletion in NA12878 by DELLY was inferred by both the reported genotype (GT) and by observing at least one high-quality variant pair (DV) in NA12878. Only GT was reported in the single dataset analyses, as GT and DV are functionally the same when requiring 4–10 read pairs of support. In both single and joint analyses using Hydra-Multi, the contribution of at least one read pair by NA12878 was required. Note: GASVPro does not simultaneously run on multiple datasets