Question

Should Xenium pseudobulk experiments be analysed like microarrays (limma) or like RNASeq (edgeR,Deseq2, or limma-voom)

0

Entering edit mode

richardallenfriedmanbrooklyn ▴ 30

@richardallenfriedmanbrooklyn-24118

Last seen 8 weeks ago

United States

Dear List,

I have xenium dataset or an experiment and control (n=3). I want to perform a differential expression pseudobulk analysis agnostic to location and cell type. Do I use a method originally designed for microarrays (limma) or one originally designed for RNASeq (edgeR, Deseq2, or voom). I would especially appreciate Gordon Smyth's input on this Issue.

Thanks and best wishes,

Rich

Richard Friedman

Hebert Irving Comprehensive Cancer Center

Columbia University Irving Medical Center

Deseq2 edgeR limma xenium Voom • 1.8k views

ADD COMMENT • link 3 months ago richardallenfriedmanbrooklyn ▴ 30

1

Entering edit mode

It's digital counts, not intensities, so naturally you would go with the RNA-seq methods.

ADD REPLY • link 3 months ago ATpoint ★ 5.0k

0

Entering edit mode

Dear ATpoint,

Thank you. That is also my understanding. A colleague thinks otherwise and specifically asked me to ask Gordon Smyth. I am asking through the forum so that there is a permanent record for anyone else who asks.

Best wishes,

Rich

ADD REPLY • link 3 months ago richardallenfriedmanbrooklyn ▴ 30

1

Entering edit mode

Would you share the rationale of your colleague for the sake of discussion?

ADD REPLY • link 3 months ago ATpoint ★ 5.0k

0

Entering edit mode

He is of the opinion that the negative binomial model is appropriate to RNASeq because of sequencing and that since Xenium uses probes it is more like microarrays. This is as best as I understand his reasoning. He suggested that I ask Gordon to resolve the disagreement.

ADD REPLY • link 3 months ago richardallenfriedmanbrooklyn ▴ 30

1

Entering edit mode

The fact that probes are used is not the deciding factor, it is how the actual gene expression is measured. Arrays hybridize probes, indeed, but they read the attached fluorescence molecules as intensities, not counts. Probes itself can and do produce counts, for example as in targeted DNA sequencing, the relatively recent 10x probe-based Flex assays, or Xenium. So it is counts, not intensities, regardless whether this is obtained by probe hybridization, mRNA capture or other methods.

ADD REPLY • link 3 months ago ATpoint ★ 5.0k

0

Entering edit mode

ATpoint,

Thank you. That is also my understanding. I will forward your reply to my colleague.

Best wishes,

Rich

ADD REPLY • link 3 months ago richardallenfriedmanbrooklyn ▴ 30

score 1 · Answer 1 · 2025-07-15

I'm not an expert in Xenium, but I agree with ATpoint's comments.

Xenium output is a file of gene detections at specified locations, each row corresponding to one RNA transcript. It's not entirely unlike RNA-seq. If you have an RNA-seq BAM file, and assign each sequence read to a gene, then you end up with a series of gene detections. The Xenium output is essentially the same except that Xenium gives the x-y coordinate for each row as well as the gene ID. If you aggregate the detections over location, then you get gene counts. So you need to use statistical methods for counts.

The pseudobulk method itself is only applicable to count-based technologies, so your question sort of answers itself.