scRNA Cell Type ID web resources

Question

[akh22]

HI,

There are literally dozens of web resources that helps you identify cell clusters generated by scRNAseq. I've uncovered and tried a few of them and there are pros and cons of each of these sites. I am curios to see which web sites are preferred or popular, and I'd appreciate if I can get a few recommendations.

Thanks in advance.

Answer 1

[jared.andrews07]

CellMarker is decent for trying to determine, well, cell markers. Has lots of tissues and cell types from both single cell and bulk studies.

PanglaoDB is similar, but lets you search with gene sets as well. It only uses single cell experiments. It also has a list of single cell papers that is updated daily that I find very useful for staying on top of the field.

SingleR is an R package that attempts to infer cell type by comparing to transcriptomes of pure cell types. Haven't used it, but it's well spoken-of.

CellHarmony is a part of a larger suite of tools called AltAnalyze that deal with scRNA data from end to end. It also infers cell type based on reference sets. I wasn't a fan of the interface or pipeline, but colleagues have said it's reasonably accurate and covers a range of tissues fairly well.

Garnett is an R package meant to work in conjunction with Monocle to allow users to create a cell type model from one data set and apply it to other data sets. I haven't used it, but it looks quite interesting and also serves as a repository for users to upload their classifier models generated by the package for others to use.

These are the most general ones I know of, though I've also had fair success for hematopoietic cells using a custom nearest-neighbor approach using data from DMAP collected from haemosphere.

[ATpoint]

SingleR became popular enough to now have a book rather than a single vignette: https://bioconductor.org/books/3.12/SingleRBook/

[jared.andrews07]

Yes, it has also been updated and added to Bioconductor since I originally created this post. It is now much more performant, has several viz options for exploring output, and has several more reference datasets available...though I'm biased as I was involved in the dev of this version.

[akh22]

I pretty much use SingleR after LTLA took over. I primarily use Seurat, and occasionally other R and Python packages for scRNAseq analysis, Single R really integrates well with these package with less headache, even with Integrated Su objects.

I wish an accompanying celldex package is updated more often. Reference databases in the celldex are pretty much immune based. Celldex (and singleR) will let you create a custom ref database, and perhaps we can share them ?

[jared.andrews07]

You can certainly submit a PR to celldex with your dataset and it will get added, though we tend to pull the data directly from a hosted source (e.g. GEO). You can also check the scRNAseq data package, which has several other datasets that will work with SingleR. Many more are also in the process of being added there as well, so you may consider using the dev version.

If you have specific datasets you'd like to see added, you can open an issue here with links to the data/metadata and it will likely be added eventually.