Hi guys,
During scRNA seq analysis, we make clusters of cells and find cluster-specific markers.
Each cluster is enriched with few types of genes and we have to annotate cell type from that.
But how we decide the exact annotation?
For example, if cluster number 1 is enriched with genes that may represent T-cells or macrophages or NKCs (using top 5-10 marker genes).
Any guidance or a link to a good method would be appreciated.
Hi there,
a similar question came up recently here. I would recommend that you have a look at the package "SingleR", which in my opinion works pretty well for cell type annotation.
You can also check out this recent benchmarking paper that looks at different cell classification methods for scRNA-seq data:
Article A comparison of automatic cell identification methods for si...
Have fun trying them out!
Thanks @Florian. I would look into the paper. I'm not a fan of reference based methods and SigleR was quite horrible already for this and one previous dataset.
But thanks for the input.
There is no generally accepted method to annotate clusters in scRNAseq, as it is a very dataset-specific problem. You should try different strategies to find out what works best for your data.
1. SingleR and similar packages, as suggested by Florian, are a good starting point and may be helpful if you need a general classification (you need to try different reference datasets in order to obtain relevant annonations).
2. Get your cluster marker genes and run some post analysis based on gene ontologies and gene enrichment tools (try ClusterProfiler by Bioconductor https://bioconductor.org/packages/release/bioc/html/clusterProfiler.html ). These may highlight pathways useful to classify your cells.
3. Search the literature for meaningful gene signatures applicable to your cellular types and score clusters based on expression (you can you the Module Score function in Seurat).
4. Download scRNAseq datasets from other authors who analyzed similar cell types, merge your data with theirs and run the necessary batch corrections to regress out confounding variables (try the Harmony package, https://github.com/immunogenomics/harmony ) and compare your clusters with the already published annotations.
5. When you design new experiments, if possible, split the cells you will use for scRNAseq and run an aliquot with the gold standard assay for classification of your cell type (is. multicolor flow cytometry, CyTOF, microscopy etc..). Depending on your expermental design, you could obviously FACS sort cells before scRNAseq or include CITE-Seq to improve cell identification.
Dear Gabriele Casirati ...
Thanks for your input, really appreciate it!
I Will work on it.
Thanks.
- Badges
- Science topic