For samples with high mean read depth, variants were called using quality-based variant detection in CLC Genomics Workbench with a minimum read depth of 40x, minimum average quality of 20, and minimum required variant count of 2 present on both the forward and the reverse reads with a strand bias interval between 20-80% [ 20, 34].
The first setting (high stringency) required variant calls on each strand with each base having a minimum coverage of 8X, a minimum base quality of 40, a strand minimum mapping quality of 40, and at least 20% of the reads to have the novel allele.
Although there is imputation uncertainty in the 1000 Genomes dataset, this is greatest for low frequency (below 1%) variants, whereas to explain away our observed epistatic interactions we would most likely require variants of higher allele frequency.
In addition, interpreting the functional effects of variants that fall outside protein-coding regions remains very difficult, and discriminating between mutations underlying cancer progression ('drivers') and changes resulting from a general decrease in genomic stability ('passengers') requires variants in many cancers to be assayed to look for those present multiple times.
Notably, turn-around time for reports requiring variant confirmation were delayed at least one week on average compared to reports with no reported variants.
Support on both strands was required and variants near gapped alignments disregarded.
For de novo rearrangement detection, we required true variant to be supported by at least four discordant reads (including multi-part alignments or 'split reads').
The minimum coverage required for variant calling was set to 5000 for all algorithms.
While 98.89% of the reference genome was covered by at least five reads (required for variant calls), 99.17% of it was covered by at least one read.
Before the reads are processed, Halvade can copy to each worker node additional files or databases that are required for variant calling.
