As stated, which one is better for differential expression analysis? When I say normal pipeline I mean limma-voom, edgeR and DESeq2 pipeline for standard analysis. Kal's z test is mentioned in this paper and I know some pre-made tools like the one on QIAgen uses this as well for single vs single comparison
The situation I have is that I have RNA-Seq data for some patients with rare translocations in leukaemia. There are two of these translocation with almost 10 or even more patients so I have no problem with them. And then there are these translocation which mostly I have only 1, some with 2 or 3 patients, for 7 different translocations.
Our hypothesis is that the two larger group can certainly form their own subtypes, and the others may fall into one of the three cases: (1) belongs to one of the two groups despite of the difference in translocation; (2) belongs to a separate group on their own i.e. complete outlier; (3) belongs to a separate group with other members in the "others" i.e. forming a third subtype. And this can be tested with clustering.
For that, I need to do supervised feature selection to filter out useless genes. I can do that with those with multiple samples, but not those isolated cases. So I thought I can get some DGE analysis and use a certain adj.p and fold change as cut off, or adj.p as ranking matrix, and select those genes for clustering.
So my main questions are: (1) when using limma/edgeR/DESeq2, should the model design include all the translocation labels in one go or should I just add the one contrast I want to make and build a new model everytime? (2) Is Kal's z test useful for one vs all DGE analysis and is it better? (3) Are there other outlier analysis working better under this situation?