CNVbd: A Method for Copy Number Variation Detection and Boundary Search

Abstract

Copy number variation (CNV) has been increasingly recognized as a type of genomic/genetic variation that plays a critical role in driving human diseases and genomic diversity. CNV detection and analysis from cancer genomes could provide crucial information for cancer diagnosis and treatment. There still remain considerable challenges in the control-free calling of CNVs accurately in cancer analysis, although advances in next-generation sequencing (NGS) technology have been inspiring the development of various computational methods. Herein, we propose a new read-depth (RD)-based approach, called CNVbd, to explore CNVs from single tumor samples of NGS data. CNVbd assembles three statistics drawn from the density peak clustering algorithm and isolation forest algorithm based on the denoised RD profile and establishes a back propagation neural network model to predict CNV bins. In addition, we designed a revision process and a boundary search algorithm to correct the false-negative predictions and refine the CNV boundaries. The performance of the proposed method is assessed on both simulation data and real sequencing datasets. The analysis shows that CNVbd is a very competitive method and can become a robust and reliable tool for analyzing CNVs in the tumor genome.