We carried out a preliminary analysis to test the ability of Birdsuite as a whole to call CNVs. CGH- or sequence-confirmed CNVs discovered using fosmid end sequencing on eight HapMap samples were used as a reference dataset18. Combining Canary calls surpassing the default confidence threshold with Birdseye calls surpassing a lod cutoff of 5, we recovered 56% of the reference CNVs that overlap at least 2 probes on the array, and 94% of those that overlap at least 20 probes. We compared these results to those from two commercial copy number analysis platforms, Nexus and Partek, using the same set of CEL files as input and default thresholds. At these settings, Nexus recovers 26% and 73% of CNVs overlapping at least 2 probes and at least 20 probes, respectively, and Partek recovers 4% and 12%. Relaxing the Nexus parameters to allow significantly more total CNV calls per genome than Birdsuite calls boosts Nexus sensitivity to 36% and 74%, still well below that of Birdsuite (Supplementary Note online).
