Discovering CNVs from read depth analysis of next generation sequencing data

Genomic Structural Variations (SV), including Copy Number Variations (CNV), are believed to contribute significantly to variations between human individuals, and may have larger effect on phenotype than SNPs [1, 2]. Therefore, the importance of their discovery has been realized. While originally detected from analysis of aCGH array data, SVs/CNVs can now be more efficiently discovered from next generation sequencing data such as Solexa/Illumina, SOLiD, Helicos etc. Here we present a novel method to detect deletions and duplications from statistical analysis of mapping density (read depth) of short reads from next generation sequencing platforms. We have adopted a mean-shift technique originally developed in computer science for image processing [3, 4] to analysis of read depth (RD) data. This approach performs the discontinuity preserving smoothing of RD signal through kernel density estimation and the mean-shift computation and results in segmentation of RD signal across genome. Then, CNVs regions are predicted from the analysis of calculated segments.