Batch effect correction between Bulk RNA-Seq and scRNA-Seq's clusters

Question

[ahmad mousavi]

Hello,

I need to compare some samples from different bulk RNA-Seq with some clusters of a scRNA-Seq (not whole sample) to find DEGs. We have undifferentiated samples performed by bulk RNA-Seq and fully differentiated samples (mature ones) are 2-3 clusters of a scRNA-Seq project. What I suppose to do is the do batch correction between them and find the DEGs.

However, I am not sure how precise would be such analysis? because they have completely different samples and have different amount of read counts. I got the raw counts and also cluster's raw read count from Seurat object.

How can I perform such analysis ? I think this would have some big problems, different in number of samples in scRNA-Seq and Bulk RNA-Seq, read count amount.

Answer 1

[ATpoint]

What I suppose to do is the do batch correction between them and find the DEGs.

You don't. It's two completely different technologies and experiments plus that batch effect is nested by the condition of interested (=fully confounded). Nothing that can be done here.

Answer 2

[LauferVA]

There are two distinct problems here.

The first is an intractable problem (so far as I understand, anyway). As ATpoint has correctly pointed out, the way you've described your data implies a problem of perfect separation. This kind of statistical problem is sinister and, so far as I know, more or less insoluble if no additional samples are available.

Now then, supposing there were no statistical problem here. In this case, I would not recommend DEG testing. Rather, I'd recommend de-convolution of the bulk RNA-seq data using the single cell data (if it were sufficient for the task) followed by joint analysis.