contrast table for 4 groups

[lw157]

Hi, I would like to make a contrast table for 4 groups, A, B, C, D. But I have trouble for the "group" in contrast_tbl.

contrasts_oi = c("'(C-D)-(A+B)','(C-A)+(D-B)'")
contrast_tbl = tibble(contrast =  c("(C-D)-(A+B)","(C-A)+(D-B)"),  group = c("C","C2"))

From the tutorial, the group must be chosen from A, B, C, D. Here I set group "C" for "(C-D)-(A+B)", but for (C-A)+(D-B), what should I set as C2 is not from ABCD.

Liuyang

[browaeysrobin]

Hi @lw157

For the moment, I recommend running this in two separate analyses (so one analysis per contrast, each with "C" as group name).

[lw157]

Hi @lw157

For the moment, I recommend running this in two separate analyses (so one analysis per contrast, each with "C" as group name).

Thanks a lot, Robin. I will run them separately with same group name.

Just one more question. I found the results (LR pairs) of "dividing by group number" is different from NOT dividing. In this case, should I divide by 4? e.g. ((C-D)-(A-B))/4 (or (C-D)/2-(A-B)/2? ), or just go with (C-D)-(A-B) ?

From multinichenetr tutorials, there is no dividing for the anti-PD1 case, but from limma manual, it seems they suggest to use "divide by group #" (page 47 from https://www.bioconductor.org/packages/release/bioc/vignettes/limma/inst/doc/usersguide.pdf ).

Thanks again.

All best, Liuyang

[browaeysrobin]

Hi @lw157

In the MultiNicheNet code of writing out your contrasts, we use the way described in 9.5.2 (without dividing when comparing equivalent groups), and not the way described in 9.5.4.

[lw157]

Thanks a lot, Robin. This is really helpful. I will close this post.