p-value digits for CellChat analysis

[toushirou1220]

Hi Authors,

I'm a postdoc from UCSF and we uses your tool a lot. It helps greatly in our daily research. We are currently making figures for our next publication. I pulled the numbers out from the CellChat object by df.net <- subsetCommunication(cellchat) However, we noticed that p-value is only calculated to 0.01. p-value < 0.01 is stored as 0. I know from a statistical point comparing p-value is not very meaningful, but we want to try plot figures like GO enrichment figures, where the GO terms (interaction pairs) are ranked by p-value and presented as horizontal bar plot. Is there a way to acquire more digits of the p-value?

Thanks, Andy

[sqjin]

@toushirou1220 I am afraid the current implement does not give more digits of the p-value due to the permutation test we used.

[PNA743]

Hi Authors, sorry for the (maybe) stupid question regarding exactly the digits: Does this mean, that a p-value of 0 is rounded because it is lower than 0.01? We were not sure in our lab if we can trust an interaction with a p-value of 0. Thanks, Susanne

[sqjin]

@PNA743 You can check our paper on the definition of the standard permutation test. P-values of 0 means none of the permutations have higher interaction probability than the observed data (i.e., the original scRNA-seq data). Thus it is not related to the threshold 0.01.

[masa187]

Hi authors,

I have a similar question to the previous one since I am not sure if I got it right from your answers in this post and I don't want to include any errors in my analysis. After getting the dataframe with only the significant L-R interactions, I have a lot of hits of L-R interactions with a p-value of 0 (as @toushirou1220 already mentioned). So is it right that p-value=0 is simply 0 because of rounding? Is the number just really small because of the permutation test, so is it actually something like 0.00000001 and not 0? Can I still use those interactions with p-value=0 for my further analysis?

Thank you,

Marijana

[sqjin]

@masa187 The interactions with p-value = 0 are definitely needed for further analysis as they are the most significant ones.

p-value=0 is simply 0. The comment p-value < 0.01 is stored as 0. is not correct.

[joan-yanqiong]

I have similar results as @masa187 , a lot of interactions with p-value = 0. I'm doing 1k permutations, and I'm wondering how likely it is that "none of the permutations have higher interaction probability than the observed data", then.