Perform Dimensionality Reduction using Graph-Regularised PCA on Spatial Data

Computes a graph-regularised PCA using spatial coordinates and scaled expression data. A k-nearest neighbour (k-NN) graph is computed using spatial locations and used to regularise the PCA decomposition via a graph Laplacian. The result are stored in the @reductions section of the returned SpaMTP Seurat object.

Usage

RunSpatialGraphPCA(
  data,
  n_components = 50,
  assay = "Spatial",
  slot = "scale.data",
  image = NULL,
  platform = "Visium",
  lambda = 0.5,
  n_neighbors = NULL,
  include_self = FALSE,
  alg = "kd_tree",
  fast = TRUE,
  graph_name = "SpatialKNN",
  reduction_name = "SpatialPCA",
  verbose = TRUE
)

Arguments

data: A SpaMTP Seurat object containing spatial data (feature data and spatial coordinates).
n_components: Integer specifying the number of principal components to compute (default = 50).
assay: Character string defining the name of the assay to use data from (default = "Spatial").
slot: Character string defining the name of the slot to extract scaled data from (default = "scale.data").
image: Character string matching the name of the image to use for extracting spatial coordinates (default = NULL).
platform: Character string matching either "Visium" or "ST" to determine how the k-NN graph is constructed. If "Visium" k-nns will handle the hexagon spot arrangement, including setting n_neighbors = 6, else "ST" assignment will set n_neighbors = 4 unless a value is specifically provided (default = "Visium").
lambda: Numeric value defining the regularisation parameter that controls the influence of the graph Laplacian (default = 0.5).
n_neighbors: Integer value specifying the number of spatial neighbours to use. If NULL, will default of 6 for "Visium" data and 4 for "ST" platforms (default = NULL).
include_self: Boolean logical value indicating whether to include self-connections in the graph (default = FALSE).
alg: Character string specifying the algorithm to use for nearest neighbour search (passed to FNN::get.knn()) (default ="kd_tree").
fast: Boolean logical value stating whether to use fast approximate eigendecomposition via RSpectra::eigs_sym(). For large datasets this is recommended (default =TRUE).
graph_name: Character string of the name to use for storing the computed spatial graph in @graphs(default ="SpatialKNN").
reduction_name: Character string stating the name of the dimensionality reduction stored in @reductions(default ="SpatialPCA").
verbose: Boolean indicating whether to show the message. If TRUE the message will be show, else the message will be suppressed (default = TRUE).

Value

A SpaMTP Seurat object with a new graph stored in @graphs and spatially-aware PCA reduction values stored in @reductions.

Details

Note: This method has been adapted from GraphPCA python package.

Examples

# spamtp_obj <- RunSpatialGraphPCA(spamtp_obj, platform = "Visium")