Good afternoon scvi team!
I am very new to the scvi tool and moslty have been working with scRNA-seq data within R environment. I came around a specific issue which scvi could potentially be well suited for so I would love to learn more and would appreciate any guidance.
I would like to adapt scVI to learn a transformation between two batches being RNA sequencing protocols. I have data where both protocols were applied to cells from the same samples (so I have “ideal” comparison pairs between batches).
Ideally, I’d like to pre-train a model on these “ideal” matched samples that could take the sample of one batch and map it into the “space” of another batch. I want to then save this model for further use on non-ideal/regular data to map it from one protocol space to the other. If this works, the “corrected/mapped” gene expression from one batch/protocol would be reconstructed to resemble that of the other batch/protocol.
Do you have recommendations or existing tooling within scvi-tools that supports this kind of use case? And if not directly supported, would it be best to subclass SCVI and adjust the loss, or is there a better entry point for this kind of modelling?
I am aware that scvi is known for its developer tools and would love to know if they would be useful/applicable in this case and how I can do so in a best way.
Thank you!