search

ventolab / CellphoneDB

CellPhoneDB can be used to search for a particular ligand/receptor, or interrogate your own HUMAN single-cell transcriptomics data.
https://www.cellphonedb.org/
MIT License
305 stars 52 forks source link

Pvalue adjusting #121

Closed maximelepetit closed 7 months ago

maximelepetit commented 1 year ago

Hi.

Is there any p-value correction steps, e.g, Bonferroni or BH, included before reporting the p-value?

Best regards !

ktroule commented 1 year ago

Hi, thanks for using CellPhoneDB.

No, P-values are uncorrected.

Regards

MarieCoutelier commented 4 months ago

Hi, We were wondering if you would recommend to correct the p-values then? And if so, how? Thanks, Marie

datasome commented 4 months ago

Hi Marie,

The authors of the package are not available to comment for the time being, but I understand an advice from a statistician was sought in the past and the recommendation was not to correct p-values. I'm afraid I'm not able to offer any more advice than this.

Best, Robert.

MarieCoutelier commented 4 months ago

Hi Robert, Thanks a lot for your input ! Do you know if the recommendation comes from the fact that the permutation method accounts for multiple testing already? All the best, Marie

datasome commented 4 months ago

Hi Marie,

It is one logical interpretation, but personally I don't see how the permutation method accounts for multiple testing - each permutation of cell types across the cells is a single test, and the more tests you do the more likely you are to encounter the result you expect. Hence imho multiple testing might be beneficial, but it's beyond my knowledge to know how many permutations warrant applying multiple testing correction. I have used https://www.statisticshowto.com/benjamini-hochberg-procedure/ in the past, but please don't treat it as an official recommendation for using with CellphoneDB analyses. Hope that helps.

Best,

Robert.

MarieCoutelier commented 3 months ago

Dear Robert,

My apologies for my late reply. I just wanted to thank you for your insightful comments, they were really useful for us to decide on an analysis strategy.

In the end, we decided to apply a BH correction per celltype-couple; it does not reduce much the amount of significant interactions, but we feel more confident with more stringent results.

All the best,

Marie

datasome commented 3 months ago

Hi Marie,

Glad to hear that you found those helpful and thank you for letting me know. I wish you best of luck in your research.

Best wishes,

Robert.