attempting batch correction with ADTnorm

[abspangler13]

Hello,

I installed ADTnorm from github using remotes::install_github("yezhengSTAT/ADTnorm", build_vignettes = FALSE)

I have an ADT-seq dataset that was generated in 4 different batches/runs with multiple samples in each batch/run. I am seeing an effect of the run even after normalizing with the DSB method. For example:

I have tried using ADTnorm with various parameters, but still the data is very separated by run. Below is my code along with the resulting UMAP plots:

Option 1:
`#### option 1
A316.VDJ <- readRDS(file = here::here("A316_final_vdj_all.rds"))
save_outpath <- "/Users/spanglerab"
run_name <- "ADTnorm_demoRun"

cell_x_adt <- t(as.data.frame(GetAssayData(A316.VDJ, assay = "Prot", slot = "counts")))
cell_x_feature <- A316.VDJ@meta.data

cell_x_feature$sample = factor(cell_x_feature$run)
cell_x_feature$batch = factor(cell_x_feature$run)

cell_x_adt_norm <- ADTnorm(
cell_x_adt = cell_x_adt,
cell_x_feature = cell_x_feature,
save_outpath = save_outpath,
study_name = run_name,
save_intermediate_fig = TRUE
)

A316.VDJ <- SetAssayData(A316.VDJ, assay="Prot",slot = "data", new.data=t(cell_x_adt_norm))

DefaultAssay(A316.VDJ) <- "Prot"
A316.VDJ <- ScaleData(A316.VDJ, features = rownames(A316.VDJ))
A316.VDJ <- RunPCA(A316.VDJ, assay = "Prot", slot = "data", features = rownames(A316.VDJ), reduction.name = "apca")
A316.VDJ <- RunUMAP(A316.VDJ, reduction = "apca", dims = 1:18, assay = "Prot", reduction.name = "prot.umap", reduction.key = "protUMAP_", n.neighbors = 40, min.dist = 0.3, local.connectivity = 3, spread = 3)
pdf(file = here::here("DimPlot_prot_UMAP_all_adt_norm.pdf"))
DimPlot(A316.VDJ, reduction = "prot.umap", label = TRUE, group.by = "run")
dev.off()`

Thanks for your help,

Abby

[yezhengSTAT]

Hello Abby,
I am not sure if you are aware of this tutorial webpage: https://yezhengstat.github.io/ADTnorm/articles/ADTnorm-tutorial.html. It contains a few typical examples in turning the other parameters. UMAP can help us diagnose if there are batch effects left in the data but not a good approach to guide us to tune the parameter. Instead, can you generate a few density plot of the protein markers and see if their peaks align before and after the ADTnorm normalization? Just like those figures in the tutorial website. From there, we can see how to tune the other parameters. ;)

Thanks,
Ye

[abspangler13]

Hi Ye,

Thanks for informing me about the tutorial page. Here are a few of the density plots from this dataset. It does appear that the peaks are aligned.

P-CCR3

P-CCR6

CD-19

Thanks for your help,

Abby

[abspangler13]

Hi Ye, I'm going to test a few more parameters and get back to you shortly. You do not need to respond to my previous comment.

[yezhengSTAT]

I was about to say that the peak alignment looks clean to me, at least for the three markers you shared with me......Do you see a big discrepancy across runs in other markers? If not, I don't see why UMAP shows a big separation across runs.......Not sure if it is the reason, but can you try skipping the "ScaleData" step? For example, you may directly run PCA on the cell_x_adt matrix using "prcomp" and then umap using "umap". Plot the umap coordinate directly on the scatter plot.

Thanks,
Ye