We were considering adding cycle consistency loss and VampPrior (see Integrating single-cell RNA-seq datasets with substantial batch effects | bioRxiv) to scvi-tools as a separate model or if possible to scVI directly as an add-on functionality. Would especially the latter be of interest? If yes, could we further discuss implementation choices as the current implementation deviates a lot from the scvi-tools backbone and is also based on an old version of it.
Hi, we’re definitely interested - could you actually submit an issue in our GitHub repo instead since this topic is more appropriate there, and then we can discuss further? Thanks!