I examined the effect of different types of filtering on the raw P value distribution for a subset of the data (randomly picked a fifth of the variables). Though I am not going to use an effect size-based filtering, it seems that such filtering does not have a very major effect in the case of my data.
