A New Genomic Wave Adjustment Method for DNA Copy Number Variation Detection

DNA copy number variations (CNV) carry information on the mis-regulation of DNA replication in cancer cells, making the study of CNVs an indispensable component of cancer genome analysis. Nevertheless, genomic waves present in various platforms of DNA copy number data can impede precise CNV detection. In this paper, we propose to enhance the classic fused Lasso algorithm by accounting for the effects of wave patterns in DNA copy number data. Specifically, we introduce a partially linear model to distinguish piecewise-constant signals from genomic waves, and employ a nonconvex penalized regression to identify multiple change-points. Moreover, our proposed method allows us to identify other potential genomic characteristics that correlate with wave patterns, and can also serve as a preprocessing step to rectify biases in other CNV calling algorithms. Simulation results demonstrate that our proposed algorithm yields more precise estimators of the number and locations of change-points when genomic waves are present. We further demonstrate the efficacy of our approach by applying it to bladder tumor aCGH data, SNP genotyping data, and breast tumor NGS data for detecting copy number variations.