Sample cell number issue

[feather-W]

Thank you very much for developing MiloR. I have a question regarding the analysis process. In MiloR, KNN neighborhoods are constructed first, and then the number of cells present in each neighborhood is counted (for each experimental sample) and used for differential abundance testing between conditions.

However, if some technical issue causes a certain cell population (e.g., plasma cells) to be particularly abundant in one sample (sample X) while being nearly absent in all other samples, and this sample X happens to belong to the treatment group, then in the MiloR analysis plasma cells would likely appear to be more abundant in the treatment group. Yet such a result would not be biological. I would like to ask whether MiloR has any correction approach to handle this kind of situation.

[MikeDMorgan]

Yes, as described in both Milo papers you can adjust for batch differences in the GL(M)M.

[feather-W]

Yes, as described in both Milo papers you can adjust for batch differences in the GL(M)M.

Thanks for reply! I wonder if I can use corrected pca(for example, harmony) to adjust batch differences? My code as follows:

seu_milo <- buildGraph(seu_milo, 
                       k = 30,d = 20,
                       reduced.dim = 'HARMONY')
seu_milo <- makeNhoods(seu_milo, 
                       prop = 0.2, k = 30, 
                       d=20, refined = TRUE,
                       reduced_dims = 'HARMONY')
seu_milo <- countCells(seu_milo, 
                       meta.data = data.frame(colData(seu_milo)), 
                       samples="orig.ident")

traj_design <- data.frame(colData(seu_milo))[,c("orig.ident", "Group")]
traj_design <- distinct(traj_design)
rownames(traj_design) <- traj_design$orig.ident
## Reorder rownames to match columns of nhoodCounts(milo)
traj_design <- traj_design[colnames(nhoodCounts(seu_milo)), , drop=FALSE]

seu_milo <- calcNhoodDistance(seu_milo, d=20,
                              reduced.dim = 'HARMONY')
contrast.all <- c("GroupRecurrent - GroupPrimary") 
model <- model.matrix(~ 0 + Group, data=traj_design)
mod.constrast <- makeContrasts(contrasts=contrast.all, levels=model)

da_results <- testNhoods(seu_milo, 
                         design = ~ 0 + Group, 
                         design.df = traj_design,
                         reduced.dim = 'HARMONY', 
                         model.contrasts = contrast.all)

da_results %>%
  arrange(- SpatialFDR) %>%
  head() 

As above, I used 'harmony' instead of pca, I wonder if it can adjust batch difference?