Hi

I would like to ask for how to perform swish with a design similar to limma blocking situation?

#  For example in the design below, the test needed is Disease vs Normal in Condition 
#  but multiple samples were taken from same individual.
#  I thought I could consider Patient as a secondary covariates, 
#  but the level is greater than 2 and swish will not run it.
   Patient Condition
1        1   Disease
2        1   Disease
3        1   Disease
4        2   Disease
5        2   Disease
6        2   Disease
7        3   Disease
8        3   Disease
9        3   Disease
10       4    Normal
11       4    Normal
12       4    Normal
13       5    Normal
14       5    Normal
15       5    Normal
16       6    Normal
17       6    Normal
18       6    Normal

Thank you very much

Swish could handle an blocked design where e.g. there were two or more batches to control for and the comparison was Disease vs Normal. So balanced with respect to condition.

Here, you have replicated samples nested within condition, which Swish cannot control for with our existing methods. The best approach for these data is limma's duplicateCorrelation. An exception would be if you had e.g. before and after samples per donor, and donors were nested within groups -- this design Swish can handle with an interaction test. But here you don't have the comparison of interest within the donor, but across donors.

So unless the samples are technical replicates (just multiple sequencing runs but not distinct samples) then I would go with limma.