| Motivation: Existing pre-processing methods for DNA microarrays designed to detect copy number variations (CNVs) lead to high false discovery rates (FDRs). High FDRs misguide researchers especially in the medical context where CNVs are wrongly associated with diseases. We propose a probabilistic latent variable model, cn.FARMS, for array-based CNV analysis which controls the FDR without loss of sensitivity. At a DNA region, cn.FARMS constructs a model by a Bayesian maximum a posteriori estimation where the unobserved, latent variable represents the copy number that is measured by observed genetic markers (probes). The latent variable’s prior prefers parameters which represent the null hypothesis, (same copy number for all samples), from which... |
