Hoping I can get some help. I have been using DESeq2 for several months now on several different projects (I manage a NGS core). For most of these projects, the samples do not cluster as I would expect. For example, not all of the treatments will cluster with the treament group and not all of the controls will cluster with the control group. Is this expected to happen sometimes or does this always indicate a problem with sample prep (sample mix-ups, contamination, ineffective treatment)? Secondly, should I always remove the samples that are not clustering with their respective group before continuing with my differential expression analysis? Thanks.

I would never remove samples because they do not cluster with their respective group. I would only remove samples if they are far separated from all the other samples, and only after I examined other properties of the samples, e.g. the FASTQC report, boxplot of the counts per gene compared to other samples, etc.

There are a couple reasons you could see groups not clustering: there is a lot of biological variability, much more than the differences between groups; there are DE genes separating the groups, but only a few, and these do not dominate the top variance genes; or there are strong batch differences and these dominate the variance in the data. This last issue can be dealt with, if the conditions are balanced within batches, by including batch or surrogate variables in the design (see workflow).

Thanks for your input Michael.