Issue with batch

[mhvbyrne]

Hi, this is a great tool but I am struggling to have the decontX function correct by batch, it runs fine without specifying batch.

For my object batch metadata is in cart_id

library(decontX)
library(celda)
library(Seurat)

sobj <- readRDS('seurat object file path')

counts <- LayerData(object = sobj, layer = "counts") 
sce <- SingleCellExperiment(list(counts = counts),
                            colData = sobj@meta.data) 
sce$cart_id <- as.character(sce$cart_id)

sce <- decontX(sce, batch='cart_id')

When I run this I get the following error: Error in fixupDN.if.valid(value, x@Dim) :
length of Dimnames[[2]] (1) is not equal to Dim[2] (22518)

I do not have any missing values for cart_id so i'm a bit confused what's wrong.

sce_list <- lapply(unique(sce$cart_id), function(batch) {
    decontX(sce[, sce$cart_id == batch])
})

# Merge the processed batches
sce <- do.call(cbind, sce_list)

I tried this which applied decontX to the separate batch samples but then the downstream analysis did not look correct (i.e. the clustering looked very odd - like it just merged all the UMAPs, and the contamination plot just looked like it took it from one sample.

I would be grateful for your help

[mhvbyrne]

I have just figured it out: sce <- decontX(sce, batch=sce$cart_id)

However, the umap figures from the vignette are still strange

Would you advise I just use the decontamination value and then do my own clustering and visualisation?

[joshua-d-campbell]

Hello, thanks for your question. Yes, that is correct the whole vector needs to be supplied as a parameter rather than the name of the column in the colData slot.

By default, the UMAP is quickly generated with the "calculateUMAP" function from the scatter package and the clusters are generated by applying dbscan to the UMAP. You can supply your own cluster labels or UMAPs though. You don't have to rely on decontX defaults. Hope that helps.

I'm moving this to a Discussion in case anyone else has the same questions.

[joshua-d-campbell]

Also, I should add that when supplying the "batch" parameter, decontX will run separately on each batch and produce separate UMAPs for each one rather than a combined one.