Even after an empirically modeled interpretation of intensity measurements, the estimate of copy number from single probes can be noisy. The next step is therefore to integrate information across neighboring probes to find strong, consistent evidence for altered copy number states. Birdseye, an HMM-based algorithm, utilizes dynamic programming to perform this search quickly and efficiently across each chromosome15 (Fig. 4). Each segment of discrete copy number is assigned a lod score indicating the relative probability of the variant versus normal copy number in the region (which can be used in downstream analyses to prioritize discovered CNVs on the basis of confidence). Because the approach uses probe-specific variances, noisier probes are inherently downweighted with respect to more responsive probes, reducing the number of false positives one would find by assuming all probes are equal (Supplementary Methods).
