21 - Copy Number Identification by Viterbi

Abstract

Copy number variations (CNVs) are known to have a causative role in a host of diseases, including cancer, and can often be used to help inform treatment decisions. For example in multiple myeloma, amplification of certain genes in chromosome 1q can lead to drug resistance. Meanwhile, deletions of 17p in multiple myeloma can lead to a loss of p53, which is predictive of a more aggressive disease. In spite of the well-known role CNVs play in human disease, highly accurate, flexible methods of identifying CNVs from targeted next generation sequencing data remains a challenge. Current algorithms generally rely on variation in sequencing depth between normal and tumor samples to identify putative targets. Unfortunately, accounting for all sources of bias associated with NGS targeted panels (capture methods, GC content, read mapping) and samples (e.g. tissue type, preservation method) which influence sequencing depth is challenging, and often results in noisy estimates of fold change. In addition, most current methods lack the ability to identify copy number neutral loss of heterozygosity events (LOH), as these events have read depths that are indistinguishable from undamaged DNA. However, all CNV events except full deletions demonstrate easily identifiable heterozygous allele frequency banding for single nucleotide variants (VAF). Here we introduce Covit (Copy number by Viterbi), a new algorithm that combines information from read depth, VAF, and a running average of homozygous allele rates by utilizing a modified Viterbi algorithm. CoVit consistently outperforms competing algorithms in identifying CNV events across tissues types and tissue matrices.