Maximum Likelihood Estimation of DNA Copy Number Variations in HR-CGH Arrays Data

摘要

We address an efficient combined maximum likelihood (ML) estimator of the copy number variations (CNVs) in the high resolution array-comparative genomic hybridization (HR-CGH) microarray measurements. The estimator implies several steps. First, sequential median smoothing is provided in order to find the candidate breakpoints and determine their number. Next, linear forward/backward (FB) smoothing is applied to each of the segments for efficient denoising. Here, we also propose a fast recursive scheme for forward and backward smoothing. Then jitter induced by the median is removed employing the ordinary least squares. Finally, constant segmental levels are assigned at the most accurate points of FB smoothing. The estimator has been tested for the HR-CGH microarray measurements simulated with the white Gaussian noise and segments having different typical copy numbers and durations.