Detecting Recurrent Copy Number Variations from aCGH Data via Smoothness-Constrained Models

Increasing number of genomic studies have associated copy number variations (CNVs) with several diseases such as cancer, autism, Alzheimer, and many autoimmune diseases. Therefore, developing reliable computational tools for recurrent CNV detection is crucial to understand the development of such diseases. A widely used microarray technology for measuring DNA copy number is array-based Comparative Genomic Hybridization (aCGH). Identifying concurrent CNVs is challenging due to the presence of noise and sample-specific variations. In this paper, we propose two matrix decomposition-based approaches, Smooth Regularization Decomposition (SRD) and Smooth Regularization Sparse Decomposition (SRSD) for reliable recurrent CNV detection from aCGH data. The essence of the two techniques is to model the aCGH profiles as smooth signals. However, the SRSD algorithm extends the SRD model to account for sample-specific variations. We also propose an algorithm to efficiently solve the SRSD model. Our simulations, using synthetic and realistic datasets, show that our proposed models achieve better accuracies when compared to the state-of-the-art models.