Fast and Robust Segmentation of Copy Number Profiles Using Multi-Scale Edge Detection

Abstract

Raw copy number data is highly dimensional, noisy and can suffer from so-called genomic wave artifacts. We introduce a novel method based on multi-scale edge detection in derivative space. By using derivatives, the algorithm was very fast and robust against genomic waves. Our method compared very well to existing state-of-the-art segmentation methods and importantly outperformed these if noise and wave artifacts were well present.