Double biological replication when using DESeq2

Question

[350769816]

Hello everyone, I'd like to perform DEG analysis on a tumor multi-region sampling dataset like Tracerx LUAD using DESeq2. Detailed experiment is illustrated in the chart below.

Multiple regions are sampled from sample tumor. Region is biological replication of patient, and patient is biological replication of condition. Now I want to perform DEG analysis between condition A and condition B. How can I specify the design matrix? Thanks

[Trivas]

Region would be a technical replicate, not a biological replicate IMO. Your design would be ~ Condition

[jkim]

Could you explain why region would be a tech replicate? I am just curious.

[350769816]

I do not really understand why region would be a technical replicate. Regions from a patient are closely correlated but they are different samples. Could you please explain? Much thanks

[ATpoint]

Yes, they're much more correlated than individual patients so you would underestimate the actual variance between samples, and that would probably lead to smaller p-values, so more DEGs, but more false positives.

Answer 1

[ATpoint]

In DESeq2 you can do ~Condition+Region but you cannot control for patient since patient is nested with Condition (1-2 are all A, 3-4 are all B). Adding Region to the design will block for the region difference. Do a PCA upfront to see whether region is a confounder worth adjusting.

You could switch to limma-voom and use duplicateCorrelation to account for the fact that the per-patient samples are more correlated than the between-patient samples. That would take care of the point raises by Trivas in terms of Region being either a biological or technical replicate.

[350769816]

I'm sorry that the table is misleading. R1 and R2 do not have any biological meaning but just for ordering.

[ATpoint]

~Condition in DESeq2 or ~Condition with Patient for duplicateCorrelation in limma.