GeoMx DSP WTA data: QC and edgeR vs limma-voom

[JoanCG]

Hi,

first of all, thanks for this very helpuful package!

I'm working on my first GeoMx transcriptomics dataset (Whole Transcriptome Atlas panel) and I am facing some doubts with QC and statistical testing for differential expression analysis.

1. My dataset contains a single probe per gene (except for negative probes). As far as I can understand, Nanostring's tutorial recommended probe QC (minProbeRatio, percentFailGrubbs and LOQ) does not apply to my dataset, and addPerROIQC should be enough. Am I correct?

2. In my dataset, I noticed that low-count genes do not exhibit greater biological coefficient of variation (BCV), which contrasts with typical patterns observed in bulk RNA-Seq data (see an example in the edgeR User's Guide). Interestingly, I observed a similar trend in the BCV plot from your "GeoMx transcriptomics with standR" tutorial. Do you have any thoughts on why this might occur? Could this affect the performance of edgeR? I personally think that GeoMx data better fits edgeR assumptions, and some experiments I carried confirmed that it captures all DEGs identified by limma-voom and some more, but this observation made me have second thoughts about using limma-voom. This is the plot I obtained with plotBCV():

Thanks a lot!