Bayesian detection of recurrent copy number alterations across multiple array samples

Copy number alterations (CNA) affecting small portions of chromosomes are difficult to identify. Advances in microarray technology now allow very high resolution scans of large cohorts of samples but at the price of severe noise degradation. Our proposed genome alteration detection algorithm (GADA) has been shown to be a highly accurate and efficient approach to analyze a single array sample. In this paper, the sparse Bayesian learning (SBL) used in GADA is extended to find CNA on multiple samples that share breakpoint positions but may have different magnitude of alteration. Our model is especially well suited to analyze sample replicates, i.e., multiple arrays from the same specimen. Our results show that replicates greatly improve the accuracy and robustness in detection. In some cases, a single replicate sample offers an accuracy equivalent to a 2-fold increase in the signal to noise ratio, while reducing by up to a 50% the detection of false CNA caused by outliers. The computational cost of the algorithm is essentially linear O(NM) in the number of the microarray probes M and samples N. In conclusion, the multiple sample GADA (N-GADA) presented here appears to be a promising tool for finely locating small CNAs that are shared across multiple samples.