For example, the read count of gene Z in condition A is 0, 0, 0, 123 and in condition B is 13000,13500,12500,14000. It is obvious that gene Z is significantly differential express between condition A and condition B, however, sometimes, DESeq2 set the pvalue to NA because of the outlier in condition A.
I suggest that before filtering the outliers, firstly testing whether the maximum value in condition A is far less than the minimum value in condition B. To do this, DESeq2 can reduce the false negative rate.
Thanks for the report. I added a fix for this in the development branch (v.1.21). It's a heuristic for the simple two group design to not filter based on Cook's distance in a case like this.