Question: SingleR for integrated scRNA-seq analysis
gravatar for mvis1231
7 months ago by
mvis123160 wrote:

Hi, I have data from a specific cell from mouse fed with a certain diet. I integrated 4 datasets that were measured at four different time for the integrated single-cell RNA seq analysis. I have been referring to the Seurat vignette :

I am using SingleR to identify cell type for each cluster and I am wondering if I need to set DefaultAssay as "RNA" or "integrated". I tried both, but they gave me slightly different results for cell type identification.

Should I keep DefaultAssay as "RNA" or "integrated"?

Any thoughts and advice are greatly appreciated. Thank you.

ADD COMMENTlink modified 7 months ago by igor12k • written 7 months ago by mvis123160
gravatar for igor
7 months ago by
United States
igor12k wrote:

The assay should be RNA, since SingleR expects expression values.

ADD COMMENTlink written 7 months ago by igor12k

Just to add => log2 expression values.

ADD REPLYlink written 7 months ago by ATpoint44k

Good point. I was just concerned with "RNA" vs "integrated".

ADD REPLYlink written 7 months ago by igor12k

But we perform integrated assay when we trying to align cell states shared across datasets. If we will use 'RNA' slot what is the point of integration?

ADD REPLYlink written 7 months ago by kinalimeric30

Integration aims to create a common clustering landscape in which all cells are embedded. This makes it easy to compare cells which (without integration) would cluster based on batch effects, biological differences such as cell cycle, all kinds of sample-specific differences. The integration values must not be used (to my knowledge) with differential analysis methods since the process creates dependencies and notably changes magnitude and direction of changes. Therefore it probably should not be used with classifiers such as singleR which operates on Spearman correlation and therefore would suffer from changes in magnitude and direction of counts.

ADD REPLYlink written 7 months ago by ATpoint44k

To think of it another way: If you are trying to identify your cells based on independent datasets, it probably makes more sense to use the stable normalized values. The integrated values will change depending on the datasets you are integrating, which means your cell types will also change and that does not seem reasonable.

ADD REPLYlink written 7 months ago by igor12k
Please log in to add an answer.


Use of this site constitutes acceptance of our User Agreement and Privacy Policy.
Powered by Biostar version 2.3.0
Traffic: 2574 users visited in the last hour