The compositional nature of scRNA-seq population data only allows statements about changes in abundance with respect to a reference group5,8,11. One way of defining such a reference is by selecting one cell type and interpreting changes to the other cell types with respect to this reference type. scCODA achieves this by forcing all effects on the reference cell type to be zero. The reference should therefore be set to a cell type that is known to be unaffected by the covariates.
