search

sqjin / CellChat

R toolkit for inference, visualization and analysis of cell-cell communication from single-cell data
GNU General Public License v3.0
614 stars 140 forks source link

Inferred interactions do not seem to match with ligand/receptor expression levels #291

Open YitengDang opened 2 years ago

YitengDang commented 2 years ago

I'm using CellChat to do a comparative analysis of two samples. I first ran a separate analysis for each sample to infer the communication probabilities and network and then merged them just like in the provided tutorial. Now it seems to me that the inferred results do not match well with the expression of ligands and receptors.

Basically, CellChat is detecting a significantly higher activity of Wnt in one condition compared to the other, but almost all ligands and receptor expression levels look similar.

Here is the inferred Wnt interaction network for both conditions: Screenshot 2021-09-20 at 15 12 53

Based on this, it detects more Wnt activity for the right-hand sample (horizontal bars): Screenshot 2021-09-20 at 15 14 00

However, the distributions of the ligands & receptors is hardly different at all! Only notable difference is Wnt4 in the last cluster (and a few Fzd expression levels), but this is a very small cluster so not necessarily representative. I also checked these distributions for more ligands / receptors in my scanpy data, and found no visible differences anywhere. Yet, CellChat is concluding that there is significantly more Wnt activity in the right-hand sample. Screenshot 2021-09-22 at 19 02 21

Finally, in the UMAPs of all pathways, Wnt is also rather close together for the two conditions (this is for "functional"), but I thought that given the rather different inferred networks above, they would be more separate? Screenshot 2021-09-20 at 15 21 49

Now I'm confused by how such an inference could have been made, and what one should conclude from the analysis in the end.

sqjin commented 2 years ago

@YitengDang You may check the tutorial on Part III: Identify the upgulated and down-regulated signaling ligand-receptor pairs to identify the other altered signaling.

CellChat is concluding that there is significantly more Wnt activity in the right-hand sample. Which function did you use for this conclusion?

YitengDang commented 2 years ago

@sqjin Thanks for the reply. I'm almost literally running the tutorial script on my own data, and this is the output.

I checked Part III, and this is actually useful for making a point. For instance, in this plot Screenshot 2021-09-22 at 18 50 54 you see an enrichment in the interaction "Wnt4 - (Fzd3+Lrp6)" for "1 (Bas) -> 2 (Sec) (R2)", but not for "1 (Bas) -> 2 (Sec) (tdTom)" (both in 3rd block). However, if you look at the expression profiles for Wnt4 in cluster 1 (Bas) , and Fzd3 and Lrp6 in cluster 2 (Sec), then they are basically indistinguishable. Therefore, I'm surprised that in one case CellChat finds a significant interaction with relatively high communication probability, whereas in the other case it is not even significant.

Therefore, even if the ligand and receptor expression profiles are almost identical by eye, CellChat can still predict a significant interaction in one case and nothing in the other case. To me, it seems like this could easily lead to misinterpretation of the results. If I show such a graph to a biologist, they will think that Wnt is massively upregulated in the right hand sample when this is not really the case.

Maybe this is not really a technical issue, but I'm questioning how one should interpret the results and what they really tell about how much cells are signalling.