Question about bar graph with rankNet

[rulunini]

Hi, CellChat developer. Hope everything is good.

I'm having trouble figuring tutorial out. https://htmlpreview.github.io/?https://github.com/sqjin/CellChat/blob/master/tutorial/Comparison_analysis_of_multiple_datasets.html Please, look at this part, 'Compare the overall information flow of each signaling pathway'.

I understand left one is relative bar graph and right one is absolute number of signaling pathway graph between NL and LS. However, when I see some genes on the left, such as CD40, VEGF, LIGHT, and CSF, it looks almost half percentage of them. Especially, the ratio of NL to LS is 25 to 75 at CSF.

So, my question is what’s meaning of the graph on the left? Plus, what is meaning of the black color of genes?

Thank you for considering my question!

[sqjin]

@rulunini The red colored ones are those more enriched in NL while blue colored ones are more enriched in LS. The Black ones did not show clear enrichment when computing a statistical test.

[liukai1029]

Hi, CellChat developer,

I am also having trouble interpreting this chart. As mentioned the Black ones did not show clear enrichment with a statistical test, however in many case, some black colored names were shown in the plot and those with black color usually have very significant p-values below the threshold.

Thank you for looking into my question!

$signaling.contribution name contribution contribution.scaled group contribution.relative.1 MIF MIF 1.2674516283 6.8613880 On_E 1.5 CLEC CLEC 0.0288302945 0.2819818 On_E 1.4 BAG BAG 0.0038351176 0.1797412 On_E 1.1 CD99 CD99 1.1758695917 6.5326988 On_E 1.0 FN1 FN1 1.8260842054 7.1900773 On_E 1.0 THBS THBS 0.4844406425 1.3797664 On_E 0.9 LAMININ LAMININ 1.2220707315 6.6970434 On_E 0.9 COLLAGEN COLLAGEN 6.1354518395 7.8474558 On_E 0.9 MK MK 0.8292580983 5.3412003 On_E 0.8 APP APP 1.0456290414 5.8753203 On_E 0.8 CXCL CXCL 1.0904862940 6.2040095 On_E 0.6 MHC-I MHC-I 8.4340795216 8.0118004 On_E 0.6 PTN PTN 0.1041994437 0.4421945 On_E 0.5 NECTIN NECTIN 0.0379752468 0.3057337 On_E 0.4 NOTCH NOTCH 0.0005235638 0.1323653 On_E 0.0 ADGRE5 ADGRE5 0.0034789223 0.1766462 On_E 0.0 ICAM ICAM 0.0415047630 0.3142730 On_E 0.0 GALECTIN GALECTIN 0.1416650931 0.5116949 On_E 0.0 MIF1 MIF 1.9050426431 7.3544219 On_NE 1.5 CLEC1 CLEC 0.0398645680 0.3103405 On_NE 1.4 BAG1 BAG 0.0042257741 0.1829306 On_NE 1.1 CD991 CD99 1.1241277797 6.3683542 On_NE 1.0 FN11 FN1 1.7419502973 7.0257327 On_NE 1.0 THBS1 THBS 0.4474354655 1.2434363 On_NE 0.9 LAMININ1 LAMININ 1.0457855966 6.0396649 On_NE 0.9 COLLAGEN1 COLLAGEN 5.3032350338 7.6831112 On_NE 0.9 MK1 MK 0.6376640769 2.2225005 On_NE 0.8 APP1 APP 0.8110773640 4.7757354 On_NE 0.8 CXCL1 CXCL 0.6582282357 2.3911801 On_NE 0.6 MHC-I1 MHC-I 5.1618715454 7.5187665 On_NE 0.6 PTN1 PTN 0.0482471307 0.3298785 On_NE 0.5 NECTIN1 NECTIN 0.0169095715 0.2451055 On_NE 0.4 NOTCH1 NOTCH 0.0000000000 0.0000000 On_NE 0.0 ADGRE51 ADGRE5 0.0000000000 0.0000000 On_NE 0.0 ICAM1 ICAM 0.0000000000 0.0000000 On_NE 0.0 GALECTIN1 GALECTIN 0.0000000000 0.0000000 On_NE 0.0 pvalues MIF 9.815523e-74 CLEC 7.251963e-04 BAG 1.316719e-03 CD99 1.240482e-13 FN1 1.524872e-58 THBS 3.865052e-14 LAMININ 5.522403e-06 COLLAGEN 3.175495e-61 MK 1.969572e-13 APP 1.449427e-63 CXCL 4.861691e-63 MHC-I 1.098117e-98 PTN 1.434017e-09 NECTIN 1.219122e-15 NOTCH 6.317043e-04 ADGRE5 3.612261e-60 ICAM 1.036924e-76 GALECTIN 1.076082e-47 MIF1 9.815523e-74 CLEC1 7.251963e-04 BAG1 1.316719e-03 CD991 1.240482e-13 FN11 1.524872e-58 THBS1 3.865052e-14 LAMININ1 5.522403e-06 COLLAGEN1 3.175495e-61 MK1 1.969572e-13 APP1 1.449427e-63 CXCL1 4.861691e-63 MHC-I1 1.098117e-98 PTN1 1.434017e-09 NECTIN1 1.219122e-15 NOTCH1 6.317043e-04 ADGRE51 3.612261e-60 ICAM1 1.036924e-76 GALECTIN1 1.076082e-47

[sqjin]

@liukai1029 This is because the overall communication strength is comparable in your case.

[liukai1029]

Dear sqjin,

First of all, thank you very much for your reply!

By "comparable", do you mean their "contribution" value is not that different (1.1758695917 vs 1.1241277797), as show in the example for CD99? I wonder if there is a threshold to determine if the two values are "comparable". Maybe there is a threshold that I missed anywhere, please advise.

e.g. name contribution contribution.scaled group contribution.relative.1 CD99 CD99 1.1758695917 6.5326988 On_E 1.0 CD991 CD99 1.1241277797 6.3683542 On_NE 1.0

@liukai1029 This is because the overall communication strength is comparable in your case.

In addition, there are cases where the contribution value are very different but scaled value are close as shown in the example below:

the contribution values are 8.697091696 vs 0.997241319 the contribution.scaled values are 459.1580534 vs 361.9918263

I wonder which value (contribution or contribution.scaled) is more representative or faithful to the difference? And what is the procedure of the scaling?

e.g. name contribution_On_E contribution_On_NE contribution.scaled_On_E contribution.scaled_On_NE contribution.relative.1_On_E contribution.relative.1_On_NE pvalues_On_E pvalues_On_NE CXCL 8.697091696 0.997241319 459.1580534 361.9918263 0.1 0.1 4.86E-63 4.86E-63

Thank you again

[JoyOtten]

Hi,

I actually have a similar question on what is more representative/faithful to the difference between these pathways either the contribution or the contribution.scaled.

Thank you

[sqjin]

@liukai1029 @JoyOtten You should check the contribution values. The contribution.scale is only used for visualization purpose.

[zysmu]

@sqjin Hi, CellChat developer, I also have trouble in using "rankNet". In "CCL", the "Relative information flow" does not match “Information flow”. In "Relative information flow", the "CCL" is similar between "response" and "notresponse". However, in “Information flow”, the "CCL" is dramatically different between "response" and "notresponse".

then, I research "gg2$data", find that the "contribution" of "CCL" is similar between two groups, but the "contribution.scale" is dramatically different (the following figure). therefore, is "Relative information flow" based in "contribution"? Is "information flow" based in "contribution.scale"?

The codes: gg1 <- rankNet(cellchat, mode = "comparison", stacked = T, sources.use = c(2,3,10,14,15,17), targets.use = c(2,4,5,6,7,8,9), comparison = c(1, 2),do.stat = T) gg2 <- rankNet(cellchat, mode = "comparison", sources.use = c(2,3,10,14,15,17), targets.use = c(2,4,5,6,7,8,9),comparison = c(1, 2),stacked = F, do.stat = T) GG2data<-gg2$data

[ManuelMoradiellos]

@zysmu I too was confused about it and found out this comment on another issue that could help you!

In my case I've opted to use the argument use.raw = TRUE for now but I'm planning on trying to slightly modify how the stacked barplot is made so both graphs can look similar by using the same values. With the scaling I could compare the results of the comparison to others that I want to make within my dataset/experiment as with the scaling the maximum values are the same (in my tests) and thus I can properly analyze the results returned by CellChat.

[apal6]

Hi,

Following up on the above questions, could you explain the relative information flow plot? I see that there are some genes which are present in just NL and some on LS condition but the middle stacked barplots are 50-50 (complement and GAS). Does that mean that gene is significant in both the conditions equally? Thank you

[JoyOtten]

Hi, from what I understand it's not significant.

[sqjin]

@apal6 @JoyOtten The plot shows the complement and GAS are comparable in both conditions and slight increase in LS conditions.

[nguyenlethuylan]

I don't really understand these numbers. Can someone explain to me why in the same pathway, two groups have different contribution and contribution.scaled values but give the same contribution.related.1?