For each copy number probe, emission probabilities are empirically estimated for an underlying hidden state of 2 copies as a normal distribution, with parameters determined by the intensities of all samples in the batch (excluding those already determined to be copy variable via Canary). Emission probabilities for a state of 0 or 1 copies are imputed using regression parameters learned from probes with known copy number variation (copy number probes on the X chromosome (males versus females), and single autosomal SNP probes (the A allele probe in AA, AB or BB samples)). Emission probabilities for extra copies are imputed assuming that the differences in like parameters between the model for each copy number state increase as a power law.
