Table 1 Summary of trained models for Dragen-based pipeline.

From: Reducing Sanger confirmation testing through false positive prediction algorithms

Variant/genotype

Best model

CV capture rate (%)

Final capture rate (%)

CV TP flag rate (%)

Final TP flag rate (%)

SNV—heterozygous

GradientBoosting

99.76 + −0.18

99.58

12.78 + −2.26

12.20

SNV—homozygous

EasyEnsemble

99.94 + −0.14

99.75

17.25 + −2.07

17.40

SNV—complex heterozygous

—

—

—

—

—

Indel—heterozygous

GradientBoosting

99.62 + −0.26

99.68

43.11 + −3.35

43.41

Indel—homozygous

GradientBoosting

99.78 + −0.27

99.50

55.65 + −4.16

55.16

Indel—complex heterozygous

GradientBoosting

99.86 + -0.14

99.60

53.45 + −5.65

54.22

  1. For each variant–genotype combination, the following table reflects the best model for our criteria, the cross-validation (CV) mean and standard deviation for capture rate and true positive (TP) flag rate, and final evaluation for capture rate and TP flag rate.
  2. SNV single-nucleotide variant.
Back to article page