Since I am using DESeq2 version 1.18.1 I observe a weird behaviour with the significant genes. As you can see on the heatmap below, the top 50 most significant genes are plotted (FDR<5%) and it is full of outliers. I am running DESeq2 with default options. I also tried to pre-filter the gene list with no success (filtering lowly expressed genes for a certain number of samples). I have 3 different groups with 3 samples each and the filtering I added is having a minimum of >10 normalized reads in at least 3 samples. I have to compute pair-wise comparisons, here a pair (2 groups):
This was not happening with the older version 1.10.1 of DESeq2. What did it change? Why is the software not able to deal with these odd situations by default?