Estimate cell types proportion by single cell expressions data and gene markers table

Question

[hazirliver]

Hi! I am trying to figure out how to analyze scRNAseq data correctly. I have a table of gene expression data for several people and manually curated tables with genes and corresponding cell types. I want to approximately define the distribution/proportion of cell types for each person. Is there any tool or pipeline which can does this?

A small subset of data is shown below.

Expression table:

Gene    Person_1    Person_2    Person_3
Gene_1  0.019662117969955   0.093297733863586   0.086448113372004
Gene_2  26.0562852234371    28.7133959061724    37.5410093212551
Gene_3  99.0450497028794    171.4762838741  149.156213981125
Gene_4  25.1942026862451    34.3968608833835    36.947878390043
Gene_5  0.120716394665295   0.704182934684665   0.361614364972839
Gene_6  11.5318044366502    45.9721688901488    35.1637713595378
Gene_7  0.836211167936764   0   0
Gene_8  24.1020178326822    21.393779772105 61.6404864301009
Gene_9  31.2439907816218    42.728726179124 147.469489096036
Gene_10 121.910937177135    130.946501595758    174.874892782585

Marker table:

Gene_1  Leukocytes
Gene_2  Leukocytes
Gene_3  Leukocytes
Gene_4  T cells
Gene_5  T cells
Gene_6  B cells
Gene_7  B cells
Gene_8  B cells
Gene_9  NK cells
Gene_10 NK cells

Answer 1

[fracarb8]

What I would do is:

1. Cluster the cells and generate a uMAP projection of your dataset using the seurat workflow
2. Annotate your cells
3. use the metadata table ([e-mail] ) to group the cells and calculate the proportions

For step 1, have a look at the detailed seurat vignettes

As you have a list of genes expressed in the cell type of interest, the annotation should be straight forward. You can either use FeaturePlot or Dotplot to see in which clusters those genes are highly expressed, or you can use the addModuleScore function to create a signature for each cell type. You can then plot the signature (e.g.FeaturePlot) and see which clusters have high signal.