ROC analysis for two-marker smoothing. The same ROC analysis as described in Figure 2 was performed here using the same set of markers (~40 K) as well as using a larger number of markers (~48 K). The ROC analysis was also performed using two-marker smoothing. In this case the smoothing was done for two random markers. If we assume that the performance of individual markers is not correlated with their position (that is, markers close together are likely to have similar performance), then this should be an accurate reflection of the resultant performance with adjacent marker smoothing. We note that at the lower false positive rate for the two-marker smoothed data, the curve is not smooth given low statistics.