[FEATURE REQUEST]: Work with large datasets about cellex

[tstannius]

Hi Brian,

Thanks for your patience.

My advice would be to try the following:

1. Ensure that your input data is loaded as float32 precision. This should reduce the memory footprint.
2. If your research question permits, filter out genes that are not of interest. To my understanding, CELLECT only uses the protein coding genes, so you could try and reduce to these, but I would recommend you check with the CELLECT devs.
3. Subsample cells from the cell groups of interest.

We are working on a version 2, which addresses the performance issues by e.g. using sparse matrices, but I expect it will not be ready for a few months.

I hope this answers your question and helps you to move on with your work.

Best,
Tobias