Next, we assessed how well a particular platform can be applied to detect copy number changes as an indication of the signal-to-noise ratio, by comparing the intensity ratios of probe-sets randomly selected from a male sample (NA10851) and a female sample (NA15510) for chromosome 2 versus chromosome X40, based on the assumption of a 2:1 ratio in males compared to females (Supplementary Methods). To estimate the sensitivity and specificity for each platform, we determined true- and false-positive rates and plotted the results as receiver operator characteristic (ROC) curves for CGH and SNP arrays (Supplementary Fig. 3a,b). The area under the curve (AUC) for the ROC analysis for each array (Table 2) shows a strong correlation with the fluorescence intensity variance as measured by derivative log2 ratio spread and interquartile range. CGH arrays generally show better signal-to-noise ratios compared to SNP arrays, probably as a consequence of longer probes on the former platform. Older arrays tend to perform less well than newer arrays from the same producer, with the exception of CNV focused arrays (Illumina 660W and Agilent 2X244K) where the
