Exome sequence read depth methods for identifying copy number changes

Abstract

Copy number variants (CNVs) play important roles in a number of human diseases and in pharmacogenetics.Powerful methods exist for CNV detection inwhole genome sequencing (WGS) data, but such data are costly to obtain.Many disease causal CNVs span or are found in genome coding regions (exons), which makes CNVdetection using whole exome sequencing (WES) data attractive. If reliably validated againstWGS-based CNVs, exome-derived CNVs have potential applications in a clinical setting. Several algorithmshavebeen developed to exploitexome data forCNVdetection and comparisonsmade to find themost suitablemethods for particular data samples.The results are not consistent across studies.Here, we review some of the exome CNVdetectionmethods based on depth of coverage profiles and examine their performance to identify problems contributing to discrepancies in published results.We also present a streamlined strategy that uses a singlemetric, the likelihood ratio, to compare exomemethods, andwe demonstrated its utility using theVarScan 2 and eXome Hidden Markov Model (XHMM) programs using paired normal and tumour exome data from chronic lymphocytic leukaemia patients.We use array-based somatic CNV (SCNV) calls as a reference standard to compute prevalence-independent statistics, such as sensitivity, specificity and likelihood ratio, for validation of the exome-derived SCNVs.We also account for factors known to influence the performance of exome read depthmethods, such as CNV size and frequency, while comparing our findingswith published results.