Full algorithm description?

[stanleyjs]

Hello,

I am attempting to understand your approach to reference mapping. I notice that at some point in your method, you are using SCTransform to compute Pearson residuals on a subset of the genes.
The reference I am using is https://www.cell.com/cell/fulltext/S0092-8674(21)00583-3.

While in the discussion of WNN for multimodal graph construction there is a brief reference to SCtransform, the Azimuth-specific methods (or citations within) do not appear to reference Pearson residuals. Therefore, it is unclear to me why and how SCT is being used. Is the SCT normalized output the input to sPCA? Are the same variable genes used from the reference as in the query?

Can you please provide a link to a complete description of Azimuth?

[AustinHartman]

Each Azimuth reference contains an SCTransform model to compute pearson residuals for a subset of genes identified as variable. When a query is annotated the first thing that happens is the reference SCTransform model is used to compute pearson residuals of the query counts on the subset of variable genes. After that, the pearson residuals in the query are used to project the query cells into the same low-dimensional PCA space as the reference