Fig. 3

Integration workflow of structural variant benchmark calls. This workflow depicts the integration process to obtain structural variant benchmark calls from 120 call sets. Numbers in the box represented remaining structural variants after each data processing step in the grey dotted boxes. Briefly, approximately 90,000 structural variants were discovered in 30 call sets of each Quartet reference sample. We first kept structural variants supported by at least two sequencing platforms or at least six pipelines from one sequencing platforms, then removed SVs with length over 10 Mb or located on centromeric or pericentromeric regions and gaps. INSs and DELs were extracted for the construction of structural variants benchmark calls. Sniffles was used to report structural variants sequences, and structural variants that failed in reporting sequences were filtered. Iris was applied to refine variant sequences. After obtaining consensus of structural variants in multiple data sets, we merged four catalogs of reproducible variants of each Quartet reference sample and obtained 31,659 SVs in total. Three genotypers were used to determine genotypes of these SVs, and only SVs with consensus genotypes in at least six of all ten genotype call sets were kept for Mendelian analysis. The final structural variant benchmark calls were shared by twins and followed Mendelian inheritance laws with parents