x-axis: true allele size — positive = expansion, negative = contraction. Gray: tool produced no genotype. Red: called hom-ref (missed the variant). Brown: called het-ref. Teal: called the wrong direction. Blue: underestimated. Green: exact match. Orange: overestimated. Purity = per-allele repeat purity (fraction of the truth allele's bases inside a perfect repeat of the motif; column RepeatPurity).
Tandem Repeat Genotyping: Tool Benchmark Sequence Accuracy (vs. the assembly truth allele sequence)
x-axis: true allele size — positive = expansion, negative = contraction. Left panel: distribution of the selected metric in each x bin (box = median and IQR). Right panel: fraction of alleles in each category of that metric. Gray: tool produced no allele sequence for this allele. Green: the best category — see per-metric meanings below. Yellow → red: increasingly wrong. Metrics:
• "edit distance (bp)" — raw Levenshtein distance between the reported and truth sequences.
Green means the bases match exactly.
• "normalized" — the same distance divided by the length of the true allele sequence, so a 5bp
error counts differently on a 40bp allele than on a 4kb one.
• "size difference (bp)" — reported allele length minus true allele length. The box plot keeps
the sign (negative = the tool called the allele too short, positive = too long), while the categories
bin its absolute value. Green means only that the length matched; the bases can still differ, so this
is a weaker claim than the edit distance.
Tandem Repeat Genotyping: Tool Benchmark Runtime, Memory, and Cost
NOTE: Memory is measured as the Maximum resident set size (RSS). Compute cost ($) on Google Cloud is measured by running the tool using the Hail Batch cloud execution system.