questions about human and zebrafish scRNAseq data integration

[GGboy-Zzz]

Hi,
I'm working on the integration of human and zebrafish scRNAseq data, refering to this benchmarking strategy and code, I got a overintegrated umap in samap analysis, yet a Unintegrated umap in scvi analysis, Is this caused by the distant homology between species？Or certain parameters may need to be adjusted？Thank you for your suggestion.
SAMAP UMAP,

SCVI UMAP,

[YY-SONG0718]

Hello! Thanks for getting in touch.

The homology mapping step required by scVI only keeps orthologous genes between the species. This step can lose too many genes for distantly related species, causing an unsuccessful integration. Humans and zebrafish are quite remote which can be tricky for scVI.

You are also right that SAMap can overintegrate. It is worth checking the river plot by SAMap or representing the cell type alignment scores as a heatmap, which can probably give you a better idea than the UMAP about the method's performance.

Alternatively, I would suggest trying scANVI if you have annotated cell types (make sure matching cell types have the same label), or try new methods such as SATURN.