PR containing the scripts used to generate the plot I tweeted at @smangul1
Also contains:
ROC-plots with QUAL.
Less impactful than I expected but roc_by_size_bin does show that there appears to be something seriously wrong with your <100bp truth set.
In addition to caller-to-truth, the script also does caller-to-caller call matching and it shows that there are many <100bp 'FP' calls called by GRIDSS, Delly, and lumpy. Are you sure your mouse truth sets comprehensively cover <100bp calls? How were they generated?
Alternate figure 2 plots
By comparing the length of each TP call to that of the truth set, one can say that BD is not in fact the most accurate caller, but is in fact, the worst of all the callers I had VCFs for.
PR containing the scripts used to generate the plot I tweeted at @smangul1
Also contains:
Less impactful than I expected but roc_by_size_bin does show that there appears to be something seriously wrong with your <100bp truth set.
In addition to caller-to-truth, the script also does caller-to-caller call matching and it shows that there are many <100bp 'FP' calls called by GRIDSS, Delly, and lumpy. Are you sure your mouse truth sets comprehensively cover <100bp calls? How were they generated?