Question about comapre cellchat among groups with different cell number or library size.

[hjh-air-cond]

Hi! First of all, Thanks for making such a good package!

Recently, I'm working on a project of multi-groups, and I found the cell number is totally different between group, like 7,000 to 30,000 cells/group.

So I run CellChat on both the orginal data and the downsampling data( downsample to the same cell number to the smallest group), And the result turn to that: The Interaction number and strength on orginal data: And on the downsampled data

Cause the 1st group on the right is only 1 sample with 6000~ cells and the 2nd group is 2 samples together with 8000~ cells total, so I choose the 2nd Group's cell number to downsample and the 1st and 2nd didn't change after downsample.

But the trends of of other groups have changed a lot after downsample, which makes me confused.

Should I downsample (or should I say normalize) the datasets when running CellChat? And if so, what should I take in to consider? ( I mean like the cell numer, nFeatures or the nCounts)? Also is there any possible and proper ways to perform the normalize?

Thanks!