This plot shows how different minimum quality score thresholds affect post-filtering accuracy for each tool. The top panel shows the overall number (and fraction) of alleles (y-axis) that didn't pass a given minimum quality score threshold (x-axis). The bottom panel shows the accuracy (y-axis) for the given threshold, where accuracy is defined as (true positive alleles + true negative alleles)/(total alleles) and computed as follows:
True positive alleles: where the tool's call exactly equals the true allele size and has a quality score above the minimum threshold
True negative alleles: where the tool's call doesn't equal the true allele size and has a quality score below the minimum threshold
Total alleles: the total number of genotyped alleles, including those above and below the quality score threshold

For GangSTR and HipSTR, the plot uses quality scores provided by the tools themselves. Since ExpansionHunter doesn't output a single quality score, the plot compares two different quality scores computed based on ExpansionHunter output:
Q from CIs = 1/exp(4 * (confidence interval size)/(total allele size)) based on the approach used by EnsembleTR
Q from reads = the fraction of reads that support the allele size vs. all allele sizes considered by ExpansionHunter at the given locus. This is computed using the information provided within ExpansionHunter's json output for each locus.

x-axis: represents the true allele size. Positive values are STR expansions, negative values are STR contractions.

White represents loci that didn't pass the selected minimum quality score (Q) threshold.
Gray represents loci where the tool didn't produce a genotype (i.e. due to errors or insufficient coverage).
Red is for loci that the tool called as homozygous reference and so entirely missed the variant.
Brown is when the tool incorrectly called an allele as being heterozygous reference.
Teal is where the tool called a contraction when the true allele is an expansion, or vice versa.
Blue is for underestimated alleles (darker blue means underestimated by more repeats).
Green represents alleles where the tool exactly matched the true allele size.
Orange is for alleles that the tool overestimated (darker orange means a larger overestimate).

NOTE: If selected, the quality score (Q) threshold is applied to the per-variant Q score repored by GangSTR and HipSTR. For ExpansionHunter, a Q score is computed using the formula from EnsemblTR: Q = 1/exp(4 * (confidence interval size)/(total allele size))