I'm working with a BD Rhabsody AbSeq data set containting a targeted scRNA and surfce protein (ADT) layer measured in two batches. In general I aim to integrate both modalities (e.g. MOFA+ or SeuratV4) for a joined analysis. While I perform QC and normalization of both data modalities independently (i.e. SCTransform or logNormalization of RNA and ADTnorm or CLR for ADT) I am wondering if there exists a current consensus if batch correction should be done prior to data set integration (i.e. independently try to correct batch for RNA e.g. using Harmony or scanorama and ADT using Harmony) or if it is better to try to model the batch effect during or after data set integration (e.g. as a grouping variable in MOFA+). I have skimed through multiple papers (e.g. https://www.cell.com/cell/pdf/S0092-8674(21)00583-3.pdf and the best practice in sc from the Theis lab) but so far I have a hard time arriving at a definitive answer (which probalby doesn't exist). But anyhow, I'm happy for any input!
