Hi,
I analyzed Xenium data of samples that were also previous analyzed by scRNAseq. In general the dataset is quite large and scRNAseq cell states were quite consistent across many (hundreds) of samples. The Xenium data used a custom panel based on the marker genes of cell types and cell states of the scRNAseq data.
While I was able to identify cell types, I was basically unable to confirm any of the cell states of the scRNAseq dataset. Most marker genes were expressed to quite similar levels across cell types, however making clear assumptions is also difficult, as reads for many genes were quite low per cell and a large fraction of reads appeared to come primarily from a limited number of highly expressed genes. I wonder how it is even possible to normalize this sparse Xenium data if you often just get 0-5 reads of a given gene in a cell. Just having 1 or 2 more reads by chance may make a significant difference. Especially smaller cells such as fibroblasts, endothelial cells or immune cell appear to have few reads in general.
In contrast, I observed clear spatial distribution of some selected genes among cells of a cell type, suggestive of a cell state, however those did not match the scRNAseq data. For example I have a state X in scRNAseq with 5 top marker genes, and 2 of those marker genes are enriched in completely different cells, while the 3 other label every cell of that cell type.
It's like analyzing the same samples with scRNAseq vs Xenium gives a comparable picture of cell types but a completely different picture of cell states, despite using the same marker genes. Is this expected? Any suggestions? Anything I am missing?
Thank you
0 answers
No answers yet.
Log in to answer this question.
-
Mixed marker expression in Xenium data
written by Song •
-
dentifying tumor vs normal cells in spatial transcriptomics
written by geno89 •
-
Query regarding single cell rna seq annotation using marker genes
written by TNF_alpha •
-
Facing problem with single cell rna seq annotation using singleR
written by TNF_alpha •
-
What is the proper way to identify the cell types of scRNAseq cluster?
written by MAPK2 •
-
How to find marker genes for whole tissue from several different sub-tissue scRNAseq datasets
written by nickeener •
-
Forum: How to deal with contamination in bulkseq samples? Looking at correlated genes?
written by Seaweed •
-
In search of Mouse limb cell type markers database
written by david.f.stein •
-
how to analyze scRNA-seq from treatment and control
written by BenHu •
-
Finding Differentially Expressed Genes Across Multiple Cell Types In A Datset
written by Bioinfo_2006
Interesting analysis, could you update your post adding these following information to give a bit more background : Which species (human or mouse, I guess) ? which organ ? Is the single cell a consecutive section of your Xenium section or something further away on the same tissue ? What is your Xenium panel, the 5k or everything is custom ?
Could you describe a bit more in detail what you consider a cell type vs. cell state ?
Xenium is not known to be a quantitative method so I am not so surprise that you are not able to retrieve the depth you are getting with your single cell. Could you give us some metrics of your Xenium dataset ? Number of probes detected, median number of reads per cell etc...
Unlike single cell, getting a read "by chance" in a cell is unlikely as the method is in situ targeting
That is why spatial information is an extra layer of information that is complementary to single cell data.
Thank you for your response. This is human liver tissue. The tissue for single cell versus Xenium are not from directly consecutive sections. They come from identical tissue blocks. For Xenium everything is a custom panel. I consider a cell type something like hepatocyte, endothelial cell, fibroblast or myeloid cell, and I consider a state different subsets of those cell types based on their gene expression and clustering if I subset the cell type from the global umap and just cluster cells of that cell type. Why is Xenium not quantitative if each measured transcripts is one event of in situ targeting, more likely to happen when more transcripts are available? Median transcripts per cell: 168 Median genes per cell: 63
What I meant by not quantitative was by comparing yield depth of Xenium comapred to in single cell. On top of that the approach is targeted with variability in probe efficiency and detection.
In your scRNAseq, have you checked what is the percentage of cells expressing a "state" specific gene in a given cell type, or does it look on all cells ? Is it present and specific enough to be detectable via Xenium ?
I looked just in a given cell type. State marker genes often show variable expression levels across certain 'states'. Other marker genes are expressed exclusively in one cell 'state'. Probes and their numbers were based on the original scRNAseq data. Xenium reads appear to be quite rare for many genes and questionable sufficient to capture expression levels (1-3 reads per cell). Some specific marker genes are also just found with 1-3 reads per cell making some of this questionable, especially if the same gene can be expressed in other cell types and transcript 'contamination' from neighboring cells can occur. I realize the limitations of the method, just expected it to look a little cleaner and similar to the scRNAseq data. Especially I expected more reads per cell in general, but this may also come from my tissue.