A good place to get immune cell deconvolution reference datasets?

Question

[RNAseqer]

Hello all,

I'm just familiarizing myself with immune cell deconvolution of differential expression data (microarray specifically) and I see that many programs require a purified reference dataset for the immune cells you are hoping to deconvolute. Looking through the various deconvolution programs it seems like the reference datasets are generated from selecting samples off of Geo or other databases.

I was wondering if there exists a standard or regularly employed reference dataset for human peripheral blood immune cells? Maybe something created to compare the various programs? Or must a reference dataset always be a custom affare created by the user for it to be applicable to their data? If so, could you recommend any guidelines for creating ones own reference set?

Finally, as an aside, I am using R 3.6 and tried to download Cellmix, but received the message that Cellmix was not available for 3.6, even though the CRAN states the requirements are R >3.0. Is Cellmix no longer supported for more recent versions of R? Is there is a version for 3.6 ?

Thanks for any help!

Answer 1

[Kevin Blighe]

For microarray, the 'Abbas' dataset, which is used by CellMix, is a typical reference. For scRNA-seq and bulk RNA-seq, methods are in active development (including one of our own from my former group: Molecular Portraits of Early Rheumatoid Arthritis Identify Clinical and Treatment Response Phenotypes).

To be quite frank about it, though, it is probably still better to merely view the expression by eye and infer cell-type that way. In scRNA-seq clusters, this can actually be fairly easy to do for trained immunologists, and even for bioinformaticians who are well versed in this area. Yes, bioinformaticians can 'do' things that wet-lab biologists and physicians / doctors do, too.

The Abbas signature is based on this published work:

• Deconvolution of Blood Microarray Data Identifies Cellular Activation Patterns in Systemic Lupus Erythematosus

The CellMix published work is:

• CellMix: a comprehensive toolbox for gene expression deconvolution

I do not believe CellMix is being maintained anymore - it was even difficult to install when I used it in 2015. It also now seems that the author's web domain is no longer online: http://web.cbio.uct.ac.za/~renaud/

Kevin

[RNAseqer]

Thank you Kevin! I'm currently looking at Cellmix as a sort of 'first pass', but if the Abbas dataset is a typical or oft used reference I'll probably try to employ it whatever software I wind up using. Cellmix does seem to be a bit finicky...

This is a microarray dataset I'm looking to wrap up work on and I was going to try several deconvolution programs and look at the degree of their agreement before picking one.

Answer 2

[jared.andrews07]

There are tons of good reference sets for immune cell types. A few of the better ones, in my experience, are the Data for Immune Cell Expression, this one from a recent Cell paper, and Haemopedia, which has several human and mouse sets.

For single-cell work, I somewhat disagree with Kevin, depending on the granularity of the cell types you're trying to identify, how prevalent dropout is of the relevant markers, and a few other factors. Many automated methods have been developed to assign cell types on both a cell and cluster basis. Bulk deconvolution, however, is a different problem entirely.

[RNAseqer]

Thank you Jared, I will look into these!