Updated scEiaD scVI model once more. We simplified pan species gene name alignment by removing "one to many" or "many to one" name alignments. All expressed genes are still retained in the individual species (if not present in the gene name name merged matrix), but we are less aggressive about merging gene names across species as we discovered some edge cases where many genes were getting merged into one and vice versa. This necessitated a new core gene <-> cell count matrix and thus a new scVI model. We also took the opportunity to add a new mouse retina development dataset from Balasubramanian et al. and a new ocular compartment dataset from Gautam et al. We leverage the new(ish) scANVI scVI approach in which we use the community cell type labels to subtly improve the scVI modelling of the cell types.
Updated diff testing model to a pseudoBulk / DESeq2 based approach. Now run on a per-species basis. Added heatmap visualization which uses the DESeq2 pseudobulk differential experession changes. Fixed bug that removed RPC from the Expression Plot view as well as display zero expression studies. Expression Plot view has a new option which flips flops the facet / x axis from Gene to whatever the cell type label is (CellType, CellType_predict, cluster).