Figure 2

The average bias (A) and log₂ variance (B) versus percentage of simulated outliers plotted for each summary method. In panel A, we see that Illumina's summary method can handle up to about 20% of saturated intensities before the bias starts to increase dramatically. The trimmed mean breaks down much earlier, at around 5%. The median is comparable to Illumina's method. Similar trends can be noticed in the variance (B).