Fig. 2

PicoCNV data de-noising. A. Raw RDR (top) and BAF (bottom) tracks for the DigiPico sample from patient 11152, derived from Illumina short read sequencing depths. RDR values were calculated in non-overlapping 1 Mb windows and corrected for GC content and mappability. BAF values were calculated at heterozygous SNPs. B. Percentage of bi-allelic wells predicted by theory (orange line) and observed for real samples (labelled crosses), as a function of the number of wells per input picograms of DNA. Theoretical predictions were obtained from a binomial distribution, assuming a random distribution of human genomes across multiple wells. Observed values were calculated as the percentage of wells containing both reference and alternative alleles at heterozygous SNPs. C. De-noised RDR and BAF tracks for the DigiPico sample from patient 11152. Coloured segments indicate copy number states fitted by PicoCNV. D. Per-patient signal-to-noise ratios (SNRs) for the RDR (left) and BAF (right). To calculate signal strength, data points in the copy number states {1,1} and {1,0} according to bulk ASCAT calls were compared. The signal was the squared difference in the means of these two sets. The noise was measured as the average within-segment variance, using segments taken from bulk ASCAT calls