Schematic overview of BreakDown. a) In the input are the bam files and a set of SV calls. Reads encompassing each SV are extracted for analysis. Genome-wide parameters such as average read count per bp per GC content are initialized. b) The encompassed reads are divided into three groups: normal, discordant and soft-clipped. Reads in each group are counted. c) Normal reads for large CNVs are counted in a series of consecutive non-overlapping bins. d) Read counts are normalized by GC contents. e) Expected read counts are defined as functions of SV, VAF and sequencing data. f) Genotype and VAF that maximize the likelihood function are derived from the expected and observed read counts. g) Variant scores are estimated that quantify the confidence of the results. Detailed explanations of the mathematics are available in Methods.