VarScan also has a minimum coverage parameter and its strong interaction with the M.A.Q. parameter is shown by Table 4. When calculating coverage, VarScan only 'counts' bases that have a quality above a given M.A.Q and thus coverage effectively decreases as M.A.Q. increases. This interaction explains why a M.A.Q. value of 39 gives the most accurate overall results for Illumina v5, however results in no SNPs being called for Illumina v4. As we set the minimum coverage to 100, any genomic position where VarScan counts less than 100 reads will be ignored. Using a M.A.Q. of 39, for the Illumina v5 simulation less than 0.003% of the genome is ignored, whereas for Illumina v4 100% of the genome is ignored. Although increasing M.A.Q. decreases false positives, there is a clear trade-off between decreasing false positives by eliminating low-quality bases and increasing false negatives by disregarding good data. This again reinforces the need to understand individual sequencing runs, even if they originate from the same technology (in our case, the sequencing runs used to create error models for the two Illumina
