Tools for bulk RNA-seq celltype deconvolution

Question

[Papyrus]

Hi all,

I've got some bulk RNA-seq data (mouse brain/hippocampus) and I'm looking for tools to extract celltype composition proportions from the data. I've been looking around and most of the tools focus on using associated single-cell data, which I do not have.

Do you know of any tool which provides celltype predictions (mouse brain) by only inputting bulk RNA-seq (e.g. using internal cell reference datasets)?

If not, is the best strategy to get some public single-cell data to use with my bulk RNA-seq (from what I've read, via MuSiC or CIBERSORTx for example)? In that case, could you recommend (if there exist) databases providing single-cell data with annotated celltypes (instead of raw, unannotated data) which I could use?

If not, what would be the best strategy to achieve this goal?

thanks for your help

Answer 1

[Friederike]

Do you know of any tool which provides celltype predictions (mouse brain) by only inputting bulk RNA-seq (e.g. using internal cell reference datasets)?

As soon as you search for tools published before 2019, the majority of deconvolution methods will probably be based on bulk RNA-seq alone. See, for example, this review from 2018.

If not, is the best strategy to get some public single-cell data to use with my bulk RNA-seq (from what I've read, via MuSiC or CIBERSORTx for example)?

Dorshizi et al, 2021 demonstrate the use of reference scRNA-seq data sets for brain-specific questions.

In that case, could you recommend (if there exist) databases providing single-cell data with annotated celltypes (instead of raw, unannotated data) which I could use?

• The Allen Brain Atlas is an incredible resource for transcriptomic (scRNA-seq) data from Mouse and Human data sets. I usually find a tad bit easier to first identify the paper with the corresponding data that I might be interested in and then track down the data on the Allen Brain Atlas servers, but I'm sure you can pick your way through.
• The scRNAseq package from bioconductor has a range of scRNA-seq data sets ready to load including cell labels
• the [SingleCell Portal[(https://singlecell.broadinstitute.org/single_cell) may be of help, too.

Apart from that, identifying publications with data sets of interest, e.g. via Pubmed or via GEO, should lead you to promising scRNAseq data set candidates; most of them should be provided with metadata (e.g. cell labels).

[Papyrus]

Thanks a lot for the comprehensive answer! I guess I was looking at too-recent methods and missed most of the previous bulk RNA-seq methods. I'll check all the resources you mention!

[Pratik]

Here's another good resource for scRNA-seq atlas data. There is brain data in there along with annotated cell types. https://descartes.brotmanbaty.org/

[Papyrus]

Thank you for the info, I'll be sure to check it out!

Answer 2

[Penny]

Hello,

I would like to do a similar thing, deconvolution of bulk rna-seq data from mouse brain cortex. The disease I am focusing is neurodegenerative. Can I ask what package/software you finally used for your analysis? Because so far I found that some popular tools e.g., Cybersort, xCell are primarily for tumor and immune cells.

I have read the article Dorshizi et al., 2021 recommended by Friederike, but if I am understanding correctly, the cellR package in the article is for human data rather than mouse.

Many thanks!

[Papyrus]

I use the R package MuSiC. Basically, you get single cell data from an external dataset, for example there are mouse brain datasets in the scRNAseq package, and you can use it to deconvolve your bulk RNA-seq data.

[Penny]

Many thanks for the instructions! Will try it.