For CNP genotyping, we used two algorithms for summarizing the data from the probe sets into a single measurement, followed by clustering the resulting measurements into discrete copy-number classes (Supplementary Information). Although the two approaches agreed on the majority of calls (genotype concordance ≥99% for 96% of common CNPs), wherever they disagreed the approach that yielded the best separated clusters for that particular CNP was preferred. The joint use of the two platforms considerably improved the separation of genotype classes (Supplementary Fig. 1).
