- 33 datasets
- 3 species (human, mouse, macaque)
- 1.2e6 cells go in, 7.7e5 come out
- 30 different cell types
- Curate data, curate cell type labels
- Re-process all with kallisto-bustools
- Grid search and scPOP to ID best integration method (winner: scVI)
- Grid search again with scVI only to ID best params (5000 HVG, 8 latent dims)
- Build scVI model on human data, then use scVI/scArches query mode to add mouse and macaque data.
- xgboost cell type prediction to label unlabelled cells
How do I add my own data to scEiaD?
It is possible! Even better, you don't have to ask me! Or tell me!