Add an additional way of investigating differential expression between clusters

[nictru]

The DGEA functionality of cellXgene works, but is pretty slow for atlas-sized datasets. It would be great if we could come up with an additional visualization option, giving the experts more flexibility in investigating the expression profiles.

[nictru]

According to this paper and Leon Hafner's experience, pseudobulk analysis should be best suited for our use case.

We could potentially group all cells from the same sample within the same cluster to one pseudo-sample and then use deseq for differential expression analysis

[nictru]

For pseudobulking, this script could be helpful

[nictru]

Also metacells could be interesting

[nictru]

This lost importance, since the performance of cellxgene is way higher since the switch to CSC sparse matrices.