-
Notifications
You must be signed in to change notification settings - Fork 32
Azimuth Reference Format
Azimuth expects there to be one SCTAssay present in the reference object that is named “refAssay”. This should have a single SCTransform model stored in the SCTModel.list slot called "refmodel"
There can be additional Assays present in the ref.Rds object if those are to be transferred to the query object (e.g. ADT).
There should be two DimReduc objects present in the reference object: refUMAP and refDR.
The refUMAP DimReduc contains the dimensional reduction information needed to project the query cells onto the reference UMAP structure. This requires that the misc slot of this DimReduc contains the UMAP model stored as an element named “model”. To generate a UMAP DimReduc with the model stored appropriately, RunUMAP should be called with the following parameter settings: umap.method = "uwot", return.model = TRUE.
The refDR DimReduc contains the reference embeddings that is used for transferring labels and imputing Assay features in the query dataset. This should contain at least 50 dimensions for use with Azimuth. It also needs to be computed/associated with the “refAssay” SCTAssay i.e. DefaultAssay(object[[“refDR”]]) should return “refAssay”.
There should be one Neighbor object containing the precomputed reference neighborhoods stored as refdr.annoy.neighbors. This needs to be computed off of the refDR DimReduc using the first 50 dimensions.
The tools slot should contain the information specifically required for use with the Azimuth app. There should be an element in tools named “AzimuthReference” (generated from the function of the same name) containing an AzimuthData object. This consists of two items:
- plotref - A DimReduc object containing the information required to generate the reference DimPlot in the app. This includes the UMAP coordinates and also a data.frame in the misc slot named “plot.metadata”, which stores and metadata that the reference umap cells could be colored by.
- colormap - A named list that contains the mapping of metadata items to colors. There should be an element here for each column in the plot.metadata. Each element is an ordered named vector where the name is the metadata class and the value is the hex code color. The order of the elements determines the order in the plotting legends.
Any metadata that is to be transferred from the reference to the query should be stored in the metadata slot as a factored variable. Additionally, there should be a column named “ori.index” which stores the mapping between the index of the cells in the object UMAP was originally run on and the cell indices in the reference object (important when using a subset of cells for the reference but the UMAP was computed on the full dataset).
This file should contain the annoy index generated when computing refdr.annoy.neighbors and will be added into the reference Seurat object when Azimuth loads the reference.
Azimuth expects there to be three assays present in the extended reference object:
-
SCTAssaynamed“refAssay”. This should have a single SCTransform model stored in theSCTModel.listslot called"refmodel" -
AssaynamedBridge. This is a dictionary representation of the scRNA-seq data so that it can map ATAC data. -
ChromatinAssaynamedATAC. This should include the ATAC data from the multimodal bridge reference.
There should be five DimReduc objects present in the reference object: refUMAP, refDR, lap, slsi, and ref.refDR.
The refUMAP DimReduc contains the dimensional reduction information needed to project the query cells onto the RNA reference UMAP structure. This requires that the misc slot of this DimReduc contains the UMAP model stored as an element named “model”. To generate a UMAP DimReduc with the model stored appropriately, RunUMAP should be called with the following parameter settings: umap.method = "uwot", return.model = TRUE.
The refDR DimReduc contains the reference embeddings that is used for transferring labels and imputing Assay features in the query dataset. This should contain at least 50 dimensions for use with Azimuth. It also needs to be computed/associated with the “refAssay” SCTAssay i.e. DefaultAssay(object[[“refDR”]]) should return “refAssay”.
The lap DimReduc contains the computed graph laplacian eigenspace for the multi-omic bridge dataset.
The slsi DimReduc contains the supervised LSI dimensional reduction computed to integrate the multimodal bridge and RNA reference.
The tools slot should contain the information specifically required for use with the Azimuth app. There should be an element in tools named “AzimuthReference” (generated from the function of the same name) containing an AzimuthData object. This consists of two items:
- plotref - A DimReduc object containing the information required to generate the reference DimPlot in the app. This includes the UMAP coordinates and also a data.frame in the misc slot named “plot.metadata”, which stores and metadata that the reference umap cells could be colored by.
- colormap - A named list that contains the mapping of metadata items to colors. There should be an element here for each column in the plot.metadata. Each element is an ordered named vector where the name is the metadata class and the value is the hex code color. The order of the elements determines the order in the plotting legends.
Any metadata that is to be transferred from the reference to the query should be stored in the metadata slot as a factored variable. Additionally, there should be a column named “ori.index” which stores the mapping between the index of the cells in the object UMAP was originally run on and the cell indices in the reference object (important when using a subset of cells for the reference but the UMAP was computed on the full dataset).