Currently, characteristic_features is disregarding negative fold changes (where a feature "changes direction") for convenience, but offering another method to quantify how "characteristic" a feature is could be useful. For instance, one could rank based on the difference between means divided by pooled variance, on the significance of a t-test or of a Mann–Whitney U test.
Currently,
characteristic_features
is disregarding negative fold changes (where a feature "changes direction") for convenience, but offering another method to quantify how "characteristic" a feature is could be useful. For instance, one could rank based on the difference between means divided by pooled variance, on the significance of a t-test or of a Mann–Whitney U test.