Fig. 2

Assessing AmpliSolve’s performance using normal samples. a Median AmpliSolve FDR (%) as a function of the model pseudo-count parameter, when using different numbers M of normal samples as training set and testing on the remaining normal samples. We consider as TP all normal variants with VAF ≥ 20% and as FP all normal variants with VAF < 20% (see Text). We consider all AmpliSolve calls that have Q-score ≥ 20. b Same as (a) when considering only AmpliSolve calls with a ‘PASS’ quality flag (see Text)