I was excited to see that scvi-model trained on the MTG reference was used for label transfer to the SEA-AD data. However, I noticed that the pretrained model is not currently available on the data and documentation page. We would like to use this model in our work, I was wondering if there is any possibility of having it uploaded to the Hugging Face scvi-hub?
My lab is particularly interested in using this pretrained model because we lack the necessary compute resources to undertake the extensive retraining required to produce a new reference model. Uploading the model would save us valuable time and resources, and also enable us to replicate and build upon the impressive work that has been done by the team.
I would be very grateful if you could consider uploading this model to the Hugging Face scvi-hub.
In response to your post and emails from others, we’ve starting looking into how we can share models that are useful to the community. Could you elaborate a bit on the methods you’d like try for applying a pre-trained model to your data (something like scArches?). Right now the models have batch covariates in them that have caused some cross-compatibility issues, but we’d be happy to train new ones that meet people’s needs and share them with the community.
Thanks for your prompt reply, and I apologize for the delay in my response. While I’m not entirely certain about the specific methods you mentioned, I’m genuinely interested in a pre-trained model that I can utilize for reference mapping, more specifically, I’m interested in using it for tasks such as cell type annotation and batch correction/integration.