Detecting haplotype amplification in cancer sequencing data

Haplotype amplification on germline variants often implies selective advantage and clonal expansion susceptibility. Efficiently and accurately estimating such amplification provides insights into tumour heterogeneity and clinical implications. Existing approaches on amplification detecting are either limited on array data or focusing on genotype level only, whose accuracy may often be hurt by read depth bias or coverage deviation in clinical sequencing. In this article, we propose HMMLD, a computational approach for haplotype amplification detection. This approach is implemented based on a hidden Markov framework. Both read depth and variant allelic frequency are modeled as observed variables, while the hidden states are mapped to unknown amplifications. As we are considering germline variants, this approach involves linkage disequilibrium and interactions between germline loci, which improves accuracy by maximizing the posterior likelihoods on observed variables. We test HMMLD, on both simulation datasets with different configurations and a set of TCGA paired sequencing samples. The results outperform the existing approaches, especially on low coverage regions.