Fig. 3: The performance of different filtering methods when applied to high-frequency variants or low-frequency variants.

The high-frequency variants and low-frequency variants used as testing data are derived from whole-genome sequencing datasets (HG001, HG003, HG004, and HG006) at 30× coverage and identified by GATK, Varscan2, Mutect2, and DeepVariant, separately. Variant calls with variant allelic frequency (VAF) of more than 20% are defined as high-frequency variants. The other variant calls are defined as low-frequency variant calls. The performance of the filtering methods is assessed on the a high frequency variants using MCC; b high frequency variants using OFO; c low frequency variants using MCC; and d low frequency variants using OFO. The circle indicates the metric score achieved by FVC when applied to each specified testing data. The error bar indicates the 95% confidence intervals (n = 4 biologically independent samples). Asterisk denotes the significance of the comparison using a one-sided paired T-test (*p < 0.05, **p < 0.001), where the null hypothesis is that the FVC performs no better than the compared method. FVC consistently shows log OFO < 0 when applied to high-frequency and low-frequency variants.