multiple samples interaction comparision
Opened this issue · 6 comments
Hi cellphonedb team,
Thank you for developing this wonderful tool to analyze cell-cell interactions.
I am new to the deg_analysis method using the new version of cellphonedb. I have encountered two problems and would like to get your help.
Question 1: Suppose I have 4 samples and I want to compare the strength of the interactions between different cell types in these samples. In this case, it seems that I should run the tutorial pipeline separately and in each run, should I calculate the DEGs for each sample or use the DEGs obtained from Seurat's integration object? I think it would be better to use the DEGs calculated in each sample as the input file, but I am still not sure.
Question 2: To compare the strength of the interaction between different samples, which value from the output file, i.e. the significant mean, should be used? Is it comparable to compare these values between different samples? should I downsample to the same total cell numbers for all four samples?
Best regards!
Hi.
If each of these 4 samples represents a different condition that is of interest for your research, you can define which genes (ligands and receptors) are specific of each condition/sample. Whether you employ the DEGs as calculated by seurat or any DEG-specific tool will depend on the question you want to answer.
You can use the means as a proxy (emphasis on proxy) to compare the interaction strength as long as your data is normalized together. Significant means will only indicate whether the interaction has been found significant or not.
Hi, I have the same question in using cellphonedb to analysis different condition scRNAseq data. Should I redefine a new celltype through "paste(celltype,condition,sep='_')", then generate DEG files for each cluster? @ktroule
Hi GGboy-Zzz,
ktroule no longer works on CellphoneDB. I'm assuming that, as in the original case above, each of your samples represents a different condition, and your objective is to find out which cell types interact with each other within each sample/condition. If that's the case, analysis of each sample/condition is a separate CellphoneDB analysis - hence I don't quite see the need for defining new 'cell types' via "paste(celltype,condition,sep='_')". IfI've misunderstood you, it would help me to understand the type of biological question you're trying to answer.
Best,
Robert.
Hello,
I have a question somehow related.
I would like to compare the strength of the interactions between different cell types in WT and HET samples. I have 3 samples for each genotype, and the number of cells in each subtype is variable among the samples. Should I run cellphoneDB for each sample individually? Or could I combine the normalized counts of the 3 samples with the same genotype and have one run for WT and one for the HET?
Thank you!
Hello,
Both can be applied; but if you apply the second option, you should consider the batch effects and you should look for some ways for batch correction on the combined counts.
Hope this helps you!
Best,
Batu
Yes, it helps! Thanks a lot for your reply!