andrewwbutler
commented
3 years ago Hi Daniil,
For the "definitely latest" image with Seurat + Azimuth, I would build off of the highest numbered satijalab/azimuth image. The satijalab/azimuth-references images mostly just have a few extra layers on top of the azimuth images to deal with reference specific data wrangling.
Dear Andrew,
Thank you very much for providing these "bleeding edge" of using Azimuth to annotate real-life datasets (like in https://github.com/satijalab/azimuth-meta-analysis).
I need to annotate human PBMC cells from COVID patients. I was hoping to replicate your lung annotation and then modify it to work on PBMCs. (By the way, can you suggest a good example to correctly annotate these "NEW COVID-specific" celltypes?)
Due to European Life Science GDPR laws I must do it in a very restrictive off-line HPC environment, where the only way to install something is to upload a singularity image. I was able to build one from your docker://satijalab/seurat:latest where I additionally install Azimuth as R package, add snakemake (to run your workflows) and add rstudio-server (to have decent IDE).
Unfortunately, there are many versions of your docker://satijalab/azimuth-references image, and your example (https://github.com/satijalab/azimuth-meta-analysis) use "azimuth-references:vitessce" rather then "azimuth-references:latest". By looking at docker definition file, docker://satijalab/azimuth-references seems to be just a docker://satijalab/seurat plus a few more layers. Is it correct? Is it safe to substitute "satijalab/azimuth-references" where the workflow asks for "satijalab/seurat" image? If not, is there a single "definitely latest" image with Seurat + Azimuth? One image will be so much easier to containerize to singularity and use on off-line cluster...
Thank you in advance, Daniil