Mann-Whitney for differential expression

Mann–Whitney $U$ test, also known as Wilcoxon rank-sum test, compare samples from two populations for nonparametric inference. In the case where two samples have roughly the same shape, we can use it as a test whether the medians differ. If, on the other hand, the shapes differ significantly, it seems possible to have visually quite distinct samples with insignificant statistic.

In this particular perturbation predicting task, a call set is constructed from performing the Benjamini-Hochberg procedure on the p-values, and a scorer generates scores using call sets from the experiment result vs prediction. Here are some of the predicted profiles.

In the case of HIST1H1B, the scorer will not call this gene. On the other hand, a Kolmogorov–Smirnov test yields a p-value of roughly $1.8\times 10^{-7}$. Similarly for Cramér–von Mises. I wonder if there is a reason behind the choice of Mann-Whitney for calling certain expressions significant or not.