I was thinking of adding this feature to Gemma itself, but it would be a much better fit in the Python package.

We provide MEX-structured TAR archives containing the single-cell data. There are endpoints to discover cell types and cell-level characteristics.

I'm planning to add continuous cell-level characteristics at some later point.

Generating those formats is a bit costly, but will be drastically improved in the next Gemma release. If it's already on disk, we have this strategy that bundles a MEX directory that is super fast.