This question is in reference to the excellent tutorial posted on bioconductor site on batch correction using scran. Specifically there is a comment that reads:
Technically, we should have performed variance modelling and feature selection after calling multiBatchNorm(). This ensures that the variance components are estimated from the same values to be used in the batch correction. In practice, this makes little difference, and it tends to be easier to process each batch separately and consolidate all results in one step as shown above.
If I'm reading this right, this means the ideal workflow is: size-factor based norm -> multi-batch norm -> size factor based norm (this must always be done before mean-variance?)-> mean-variance trend (gene selection) -> MNN?