Implement in scvi-tools

[adamgayoso]

Hello,

I found your manuscript to be interesting and I'm wondering whether you have any interest in implementing a version that takes a pre-trained scvi-tools model as input (e.g., scVI) . I think this would get a lot of usage in our package!

[jaydu1]

Hi,

Thanks for the great proposal!

We're currently finishing and revising the paper to include more real data applications.
A few updates will also be made to the model, to make it more robust and stable. I expect to spend about one month on finishing this.

And yes, after we settle down on the model somehow, we would be very happy to implement a version with scvi-tools as well! Though I'm not sure if it is possible to use a pre-train model from scvi-tools, add new layers to the latent space, and continue training both of them? Or if you have any reference on similar usage of your package, that would be helpful.

[adamgayoso]

Though I'm not sure if it is possible to use a pre-train model from scvi-tools, add new layers to the latent space, and continue training both of them? Or if you have any reference on similar usage of your package, that would be helpful.

This is definitely possible, we do this for scANVI which is warm-started with a simpler scvi model (see here)

[shahrozeabbas]

Hi, just came across this tool and the preprint. Commenting to offer a vote in favor of a scvi-tools implementation. Looking forward to the final paper!