Hey @cane11 ,
We have been using baysor followed by scVI so far for batch correction on Xenium datasets. It’s working fine out of the box where it’s able to discern major cell populations. One problem we see frequently is the mixtures of mis-segmented cells (T cells with macrophages for example). Is there any way of changing the model parameters for scVI that can help mitigate the issue?
Thanks
Hi, I reacted to this privately. In the mean time, resolVI is released that addresses this and other problems with ST data.
Hey @cane11 , I got access to it but haven’t tried it yet. We just ended up using scVI for our current dataset. Is the run time of resolVI similar to scVI? Thank you for the access.
Hey @cane11. Is there a way to set batch for spatial coordinates differently than for batch correction for resolVI? My issue is I have individual TMA cores that I want to compute spatial neighbors on but I want to use TMA slides as a batch for batch correction.
Hi, it’s quite straightforward to change the code to allow for it. However, are your TMA not spaced in a manner that you anyhow only get neighbors within each TMA and not across?
You can also set the data up manually and skip the resolVI step: scvi-tools/src/scvi/external/resolvi/_model.py at 9396c33f82193e8e23ddbfbf27921d7fd2972078 · scverse/scvi-tools · GitHub
Yeah, I ended up copying the _prepare_data function and change it myself. When you say TMA, I assume you mean individual punches (cores) that are far enough on a single TMA slide; yes they are for the most part but there might be some edge effects or just cores that end up getting squished together.
Also, I see that the parameters in the reproducibility repository aren’t the default ones in resolVI. Do you prefer one over the other?