Normalization for Dataset with Multiple Conditions and Multiple Samples in Each Condition

[Nack3l]

I have a dataset containing multiomodal (RNA + ADT) data that has 9 samples across three different batches, with each batch loading 3, multiplexed samples. Batch 1 contains 3 samples corresponding to three healthy controls; batch 2 contains 3 samples taken from a single patient at 3 different time points (0, 3, 12 months); batch 3 contains 3 samples taken from another patient at 3 different time points (0, 3, 12 months). This dataset is a bit unique because while each batch contains 3 samples, batch 1 contains 3 unique replicates, while batch 2 and 3 contain only a single replicate at 3 time points.

I currently have a Seurat object that merges all 9 samples after standard demultiplexing and QC, and I understand that it is most appropriate to split my assays into multiple layers before I perform normalization (SCT for RNA and CLR for ADT). However, I am unsure as to whether I should split my object into 9 layers, with each layer corresponding to a sample, or 3 layers, with each layer corresonding to a batch. In other words, should I normalize each sample individually or each batch individually? I've currently gone with the former option, so my code looks as follows:

#Splitting merged Seurat object by sample (not batch), generating 9 different lyaers
comb.nonorm2[["RNA"]] <- split(comb.nonorm2[["RNA"]], f = comb.nonorm2$sample)
comb.nonorm2[["ADT"]] <- split(comb.nonorm2[["ADT"]], f = comb.nonorm2$sample)

#Normalizing both RNA and ADT by layer, so normalization is performed individually for each of the 9 layers
comb.nonorm2 <- NormalizeData(comb.nonorm2, assay = "ADT", normalization.method = "CLR")
comb.nonorm2 <- SCTransform(comb.nonorm2)

#Performing integration
comb.nonorm3 <- IntegrateLayers(object = comb.nonorm3, method = CCAIntegration, normalization.method = "SCT", verbose = F)