Hello everyone,
I am using a lot the high quality models provided in the scvi ecosystem. I am using scvi models for various tasks: deconvolution, integration, dimensionality reduction …
However, I am not sure to what extent these models can be used on VisiumHD data, since VisiumHD is very different from Visium V2: high number of observations (~700 000 for 8um bins), and low number of counts per bin (~50-100 for 8um bins). I am not an expert, but this might impact both the statistical models of variational autoencoding bayes models, and the way to train/retrain the model.
So, if someone has already thought about how to adapt models designed for Visium data to VisiumHD data, do you have a set of recommandations regarding statistical modelling/model training that you could share ?
Thank you very much,
Alexis