Document cell type annotations

[allyhawkins]

We will need to update both the processing information doc and the SCE contents documents to include information about the available cell type annotations.

[sjspielman]

Some details about what belongs here:

SCE contents

Noting some of was captured in #133, which I'm going to close in favor of collecting everything in this issue.

If SingleR was run...

• metadata in the processed SCE:
  • singler_results is the full SingleR object
  • singler_reference_name is the reference dataset name
  • singler_reference_label is the label type that was used in the reference (generally label.ont)
  • The celltype_methods vector will contain "singler"
• colData in the processed SCE:
  • singler_celltype_annotation are the annotated cell types, taken from pruned labels (so unclassified are NA)
  • singler_celltype_ontology is the ontology for the cell type. Only present if ontologies were used for cell typing in the first place

If CellAssign was run...

• metadata in the processed SCE:
  • cellassign_predictions are matrix of all predictions; full result analogous to singler_results
  • cellassign_reference is the organ for which marker genes were obtained from PanglaoDB
  • The celltype_methods vector will contain "cellassign"
• colData in the processed SCE:
  • cellassign_celltype_annotation are the annotated cell types where unclassified are labeled as "other"
  • cellassign_max_prediction is the prediction score, which is a probability of the assigned celltype given the provided cell types

[sjspielman]

More details!

SingleR implementation

• We identify a reference based on the tissue type of the given sample, generally using the celldex package.
• We consider reference ontology labels when possible
• We save the pruned.labels to the colData

CellAssign implementation

• We identify a set of marker genes using PanglaoDB; for given tissue type we use all genes associated with that organ
• We include an additional cell type "other" which has 0 expression of the marker genes so that cells which cannot be confidently annotated based on the set of marker genes end up getting annotated as "other"

[sjspielman]

Closed by #134