search

theislab / scCODA

A Bayesian model for compositional single-cell data analysis
BSD 3-Clause "New" or "Revised" License
147 stars 24 forks source link

how to assess sccoda results #34

Closed FADHLyemen closed 2 years ago

FADHLyemen commented 3 years ago

Hi, Sccoda is suggesting some cell types to be affected by the inhibitor, what further analysis to perform on sccoda results?

johannesostner commented 3 years ago

Hi! From there on, there's not much more to be done - scCODA simply tells you which cell types are differentially abundant with respect to the reference cell type you chose. The sign of the "Final parameter" value tells you whether there is an increase (positive sign) or decrease (negative sign) happening.

FADHLyemen commented 3 years ago

I have some suggestions:

  1. do DEGs and see if a lot of genes are expressed in your condition compare to the reference.

  2. plot PAGA and see if sccoda suggested cells connected to the other cells at the most common left "this cell trigger other cells"

  3. plot velocity and see if flow comes from your sccoda suggested cell types to the other cell types

mbuttner commented 3 years ago

Hi,

thank you very much for your comments. We designed scCODA to detect differential abundance of cell types across conditions. These conditions may be, amongst others, a control - disease setting or different time points in a differentiation process. scCODA allows identifying which cell types are actually changing across condition, however, the interpretation of the result is strongly connected to your study case.

  1. comparing the differentially expressed genes in cell types across conditions is a good idea. However, when I think of a control - disease setting, we may see a lot of genes changing due to disease response (for instance, increase in inflammation). DE genes may give an insight on the molecular drivers for the disease, but do not tell whether the shift in abundance is caused by the change in gene expression level (unless you observe a distinct change in proliferation, but that's rarely the case). Therefore, scCODA complements differential expression analysis.

  2. and 3. As far as I understand your suggestions, you have a differentiation process in mind. Indeed, there has been evidence on genetic blocks in differentiation, when certain genes are knocked out (for context, please have a look at Sam Morrison's work on the model CAPYBARA). The reason why a cell type is lost in a knock-out might be related to a block in differentiation of a progenitor (which in turn may accumulate). Another option would be an increase in proliferation of a progenitor leading to an increase of a downstream cell type (assuming that the differentiation happens at the same rate as in a control setting, without environmentally driven feedback mechanisms).

I think that your suggestions are quite interesting and I believe that scCODA provides complementary insights apart from gene expression pattern analysis across conditions. Given the various applications of single-cell RNA-sequencing, it is next to impossible to give general recommendations, though. We rather recommend to critically validate findings also with complementary experimental methods, such as cell sorting and staining, wherever possible.