Fig. 3: Assessment of optimal coverage for WGS.

a Barplots demonstrating sensitivity of variant detection and 95% confidence intervals (error bars) by coverage depth (100x, 80x, 60x, and 30–40x) from left to right for: 1. clinically relevant events detected by MSK-IMPACT and WGS (n = 220), 2. genome-wide SNVs, 3. genome-wide indels, and 4. genome-wide SVs. Only data from samples with original median coverage >100x (n = 32) are shown. Red dots indicate overall sensitivity of all mutations across all BAMs at the same subsampling level. b Histograms of variant allele frequencies for each subsampling level for a representative sample in the study cohort (H135973), showing loss in sensitivity to detect subclonal mutations at lower sequencing depth of coverage. c Scatterplot of effective local coverage vs VAF in subsampled BAMs for the clinically relevant calls from MSK-IMPACT. Variants called in subsampled BAMs are shown with circles, while the missed variants are denoted with X’s. Trendline shows the cumulative binomial distribution for obtaining at least 2 variant reads, given the effective coverage and variant allele fraction. Source data for panels a, c are provided at the data repository. Raw data for panel b can be accessed at the dbGAP study.