Statistical test on Single-cell clusters(cell types)

Hi All,

I have identified 5 different cell types from a single cell cluster, we have also identified a number of cells present in each condition(i.e. in Normal and Tumor clusters). I wanted to conduct statistical testing of differences in Gini coefficients between tumour and non-malignant compartments across all phenotypic clusters using exact permutation testing. I am unable to conduct this test, Please help me in conducting this test on my cell types. how best i can do this? i am very confused.

Thanks a lot for your kind help.

Stay safe, Dave.

enter image description here

statistical single cell exact permutation test

• 3,354 views • link

updated 4.3 years ago by Kevin Blighe

• written 6.0 years ago by David_emir

1 answer

Hey David / Dave,

Please clarify:

you say 5 cell-types but there are 15 in your table
what values are these in your table?
where are your Gini coefficients? - another metric, like mean decrease accuracy, may be more readily-interpretable.
how did you produce the Gini coefficients?

As usual, please show all code that you have used.

• 0 views • link

written 6.0 years ago by Kevin Blighe

Hi Kevin, Sorry for that not so clear question. I have 5-cell types identified (the excell snapshot is just for the illustration) - B-cell, Myeloid cell, T-cell, Epithelial, Stromal cell. and according to this paper they conducted Statistical comparisons of Gini coefficients across compartments using Wilcoxon signed-rank test with Benjamini-Hochberg correction for multiple testing; statistical testing of differences in Gini coefficients between tumour and non-malignant compartments across all phenotypic clusters was also done by the authors using exact permutation testing. I wanted to conduct the same analysis on my cell types, but I was clueless on how to apply the same test on my data. Can you please help me? The values I am considering is the cell counts(No of cells) in Tumor and normal. Thanks, Dave

• 0 views • link

written 6.0 years ago by David_emir

It is not readily clear what they did. A Wilcoxon signed rank exact test can be easily performed in R like this:

a <- rexp(1000, rate=.1)
b <- rexp(1000, rate=.1)

wilcox.test(a,b,paired=TRUE, exact=TRUE)

        Wilcoxon signed rank exact test

data:  a and b
V = 248875, p-value = 0.8805
alternative hypothesis: true location shift is not equal to 0

This test is already doing permutation in order to calculate the exact p-value. If you are unsure, I suggest posting on a a forum dedicated to statistics, or contact the authors of the published work.

• 0 views • link

written 6.0 years ago by Kevin Blighe

Thanks a lot for your help Kevin, so in. my case the inputs would be no of cells in a particular cell type , say CD4-TREG cell count in Normal would be my a=423 ?

• 0 views • link

written 6.0 years ago by David_emir

I thought that they were comparing Gini coefficients?

• 0 views • link

written 6.0 years ago by Kevin Blighe

why are you doing a paired test here? shouldn't it be the wilcoxon rank sum test instead? you have two independent groups of cells per condition within the cell type cluster.

• 0 views • link

written 4.3 years ago by gt •

Dear gt, David_emir said:

I have 5-cell types identified (the excell snapshot is just for the illustration) - B-cell, Myeloid cell, T-cell, Epithelial, Stromal cell. and according to this paper they conducted Statistical comparisons of Gini coefficients across compartments using Wilcoxon signed-rank test with Benjamini-Hochberg correction for multiple testing;

So, I said:

It is not readily clear what they did. A Wilcoxon signed rank exact test can be easily performed in R like this:

Thus, for clarity, you may ask a question to the authors.

• 0 views • link

written 4.3 years ago by Kevin Blighe