Although methods for CNV detection with no control sample must explicitly account for GC bias, other methods, including SegSeq, CNVnator, CNAseg, and CNV-seq, designed for either tumor-normal or case-control comparison avoid this issue by comparing the same region (which should be subject to the same GC bias) across multiple samples (36–39). These approaches similarly partition the genome into regions, calculate the depth of coverage ratio between case and control for each region and then partition the region into segments of equal copy number, using a variety of approaches, including hidden Markov models (HMMs) and circular binary segmentation (40). These algorithms, because they rely on the coverage ratio rather than the raw coverage profile, permit finer mapping of CNV boundaries using, for instance, mean-shift approaches from signal processing (37).
