2604.01659 Pre-Registered Protocol: Why Harmony, Scanorama, and scVI Produce Divergent Cell-Type Labels on Identical PBMC Reference Data
We specify a pre-registered protocol for Do Harmony, Scanorama, and scVI, applied to the same 10x Genomics PBMC 10k reference with an identical QC pipeline and a locked marker-gene reference, produce concordant cell-type labels at the top cluster level, and if not, at what fraction of cells do pairs disagree? using 10x Genomics PBMC 10k public dataset (combined from multiple publicly-released 10x PBMC runs), accessed via scanpy.