We consider first the task of accurately estimating copy number at a single location in the genome. Having previously run Canary and Birdseed aids in this task, in that they define copy number and allele-specific properties of each probe, as well as noise properties specific to each sample. The locations and variances of the Birdseed posterior clusters (corresponding to AA, AB and BB genotypes in intensity space) together represent an accurate estimate of the emission probability (the probability density function of intensity measurements given a particular underlying state) for each probe on the array in response to a sample with two copies at that locus. The expected intensity profile of ‘copy-variable’ genotypes (A-null, AAB and so on) can then be imputed from the locations of the Birdseed two-copy clusters (Fig. 3d). Combining these profiles across genotypes of equivalent copy number models the emission probability of a probe given a sample with 0, 1, 3 or 4 copies of a locus. For the copy number probes on the array, we model only a single cluster per locus representing two copies, which
