FindsmNN: Find Single-Modality Nearest Neighbors

Description

A wrapper function that builds a Neighbor object from a single reduction (e.g. the PLS embedding), mapping every cell to its nearest landmark cells.

Usage

FindsmNN(
  object,
  sketch.assay = NULL,
  reduction,
  dims = NULL,
  k.nn = 20,
  l2.norm = TRUE,
  nn.name = "single.nn",
  verbose = TRUE,
  ...
)

Value

return a Seurat object that contains a Neighbor object between the landmark cells and all the other cells.

Arguments

object: A Seurat object
sketch.assay: The name of the sketched assay of the landmark cells.
reduction: The name of a single dimensional reduction to use (e.g. "proj.pls").
dims: The dimensions of the reduction to use. If NULL (default), all dimensions are used.
k.nn: The number of nearest neighbors to compute for each cell. Honored as given (unlike FindmmNN, which clamps to a minimum of 20).
l2.norm: Perform L2 normalization on the cell embeddings
nn.name: Name under which the resulting Neighbor object is stored.
verbose: Print progress bars and output
...: Arguments passed to Seurat::FindNeighbors

Details

This is the single-modality counterpart of FindmmNN: instead of combining several reductions with WNN weights, it runs a single Seurat::FindNeighbors call on one reduction, with the landmark (sketched) cells as the reference and all cells as the query. The resulting Neighbor object is a drop-in replacement for the weighted.nn produced by FindmmNN and is consumed by GenerateSampleObject in the same way.