Enhanced Bayesian detection for copy number alterations from next-generation sequencing data

Next-generation sequencing (NGS) promises highresolution landscapes of cancer genomes, especially the genomewide copy number alterations (CNAs). Detecting CNAs from tumor NGS data often encounters several critical issues that heavily affect the accuracy of the results. For instance, the over-dispersed distribution of read depth signals gives rise to a challenge of designing appropriate statistical model to effectively explain the data; aneuploidy of tumor genome often causes ambiguity in interpreting copy number status for a genomic fragment. We introduce a new bioinformatics tool called EBHMM to accurately infer CNAs from single tumor sample. EBHMM employs a novel hidden Markov model (HMM) to jointly analyze read depth and read counts signals derived from tumor NGS data. To improve CNA detection robustness against severe signal fluctuation and distribution shift of read counts induced by aneuploidy, EBHMM exploits more appropriate emission models in the HMM and prior knowledge to enable accurate inference. Experimental results on real data suggest the proposed method is highly effective in detecting CNAs as well as genome ploidy, and achieves comparable performance to the state-of-the-art methods. The EBHMM software is freely available at https://github.com/qasimyu/ebhmm.