Fig. 4.

Two traditional filtering cutoffs. a For datasets with more than one thousand features, MetaboAnalyst recommends removing the lowest 40% of features according to average abundance [12]. The vertical line shown here is an example of such a cutoff, with all features to the left of the vertical line being removed. With this cutoff, many features with higher average abundance in blank samples (below the blue line) are still retained. b Some filtering methods use a pre-specified cutoff for the ratio or difference between average abundance in blank versus biological samples [10]. The black horizontal line shown here represents such a cutoff. All features with a difference in average abundance between biological and blank samples less than two (below the black line) would be removed. If such a traditional filtering approach is to be used, the MD-plot can, at the very least, help users to identify a more appropriate, less arbitrary cutoff for their dataset that strikes a better balance between removing low quality features and retaining high quality ones