Hi,

I am analyzing NanoString transcriptome data. The design of my study is as follows:

Each subject has 1 measurement only (cell line), however 2 subjects have two measurements (cell lines). I have 21 samples in total that corresponds to 19 subjects and there are 2 types of cell lines. Basically, trying to identify differentially exp. genes between two cell lines. I was able to run lmFit, Does negative corfit$consensus cause issues while fitting a model in limma results?

corfit <- duplicateCorrelation(data,design,block=Sample_metadata$`Subject ID`)
corfit$consensus 
[1] -0.5054192

Thank you,

Toufiq ```

Yes, it does cause an issue. Since the correlation is negative, you should remove blocking from the linear model. Estimating correlations from just two repeated subjects is pretty hard anyway.

I would also suggest trouble-shooting your data, because my guess is that one of the four samples belonging to the two subjects with repeat cell lines is an outlier.