Fig. 6

Distributions of estimated ICC values for high and low quality peaks in the training set for the CRC data. We use (a) box plots and (b) density plots to visualize the modes and percentiles of the distributions of the ICC values for high and low quality features in the training set, and choose a filtering cutoff of 0.43. The high quality features have a mode close to one, and the distribution of the low quality features has a larger left tail, suggesting that a cutoff to the left of the mode of the high quality features would be appropriate