Open vjcitn opened 1 year ago
Hey Vince, sorry for the delayed response. Really nice to hear that you're working on pushing this effort. With regards tot he pretrained models, we have not found a good way of automating the model training and publishing of pretrained weights with new versions yet given that the resource requirements would be too high for anything CI. We have therefore focused on fixing bug and curating more datasets for the versions since the release of the paper. It would definitely be good to think about a semi-automated way to update the weights too with every major version release.
I have been considering how to interface to Bioconductor -- here's a very early effort: https://vjcitn.github.io/BiocSfaira/articles/BiocSfaira.html
I have looked around the portal, Genome Biol paper etc. but do not see a clear description of how the various embedding and celltyping resources are produced or are evolving through the different version numbers. Am I missing a link, or can more background be provided here?
Thanks.