Hi to all,

I have an RNA-seq study experimental design where I am testing for differential expression (DE) between two groups of individuals, though I have multiple biological samples per individual and an unbalanced number of samples per individual. I am currently only interested in finding DE genes between the two groups as a whole and not any particular questions at the sample level.

These are biopsies taken from an individual at the same time and from the same tissue type at different locations. Some individuals have more samples than others. Here is an example design matrix:

     Subject  Group  Biopsy
 1   1        NR     1
 2   1        NR     2
 3   2        R      1
 4   2        R      2
 5   2        R      3
 6   3        NR     1
 7   4        R      1
 8   4        R      2
 9   5        NR     1
10   5        NR     2
11   5        NR     3
12   5        NR     4
13   6        R      1
14   6        R      2
15   6        R      3
16   7        NR     1    

I've searched online and read through a few similar questions including the relevant sections in the edgeR, limma, and DESeq2 user guides. So far either the design or DE question I've seen is slightly different than mine, such as asking questions about DE at the sample level as well as between groups, or the design having the exact same number of samples per individual.

Since the samples within an individual are expected to be correlated, does the proper DE analysis between groups need to follow a repeated measures (with duplicateCorrelation) or block design ~Subject + Group? Or am I overthinking this and it's simply still a straightforward two-group design ~Group? Could the fact that some individuals have more samples than others potentially skew an ordinary two-group comparison that should be handled in some way?

The limma User's Guide Section 9.7. Multi-level Experiments and edgeR User's Guide Section 3.5 Comparisons both between and within subjects have a similar design though exactly two samples per subject. It was suggested in the limma guide that only looking at DE between groups is a two-group comparison, "If we only wanted to compared diseased to normal, we could do an ordinary two group comparison.". Does this still apply to unbalanced number of samples per subject?

As you have already noted, this is a straightforward application of a limma two-group comparison with design <- model.matrix(~Group) and block=Subject. The fact that there is a differing number of samples per subject is not a problem. Note that you can't include Subject in the design matrix -- that would make it impossible to test the group effect.

You have labelled the Biopsies from A to D, but I assume there is no meaning to the letters. I assume you simply have one or more biopsies for each subject.