CNVScan: detecting borderline copy number variations in NGS data via scan statistics

In this paper we propose CNVScan, a CNV detection method based on scan statistics that overcomes limitations of previous read count (RC) based approaches mainly by being a window-less approach. The scans statistics have been used before mainly in epidemiology and ecology studies, but never before was applied to the CNV detection problem to the best of our knowledge. Since we avoid windowing we do not have to choose an optimal window-size which is a key step in many previous approaches. Extensive simulated experiments with single read data in extreme situations (low coverage, short reads, homo/heterozygoticity) show that this approach is very effective for a range of small CNV (200-500 bp) for which previous state-of-the-art methods are not suitable.