Figure 4

Overlay of concentration ellipses for the bivariate standard normal on real data with prominent outliers removed. Scatter plots of nine datasets from Hughes et al. [18] after outlier removal with concentration ellipses for the standard bivariate normal distribution (see Figure 1 for the portions captured). Two-sided 99.9% cut-off and robust measure of scale (median absolute deviation) for residuals were used to identify outliers. Channel intensity values were log(base 2)-transformed, normalized and standardized.