Hello, I would like to try edgeR 4 QL-methods to analyze (non-sequencing) gene count data derived from the nanostring CosMx platform and am writing to ask advice on setting up the design for the following situation, illustrated by the attached figure For each of 4 patients, I have a varying number of Regions Of Interest (ROIs) and the comparison I am interested in is between the inner and outer zones of these regions. The ROIs are therefore biological replicates in turn nested within the patient biological replicates. If there was no Patient factor I would set it up as a simple paired design.

design <- model.matrix(~ ROI + zone)

How do I set up the design so to account for correlation betweein ROIs from the same patient? Thank you

You can still use model.matrix(~ ROI + zone). There is no need to account for correlations within patient because the comparison is entirely within ROI pairs and because the ROI baselines fully account for all differences between patients as well as between ROI pairs.

This is a general rule. If a blocking factor in the design matrix is entirely nested within a coarser factor, you only need to include the lowest-level blocking factor.