Skip to contents

Calculate the local auto-correlation of a gene

hotspot_autocor_sc.Rd

This method implements the HotSpot approach (see DeTomaso, et al.) to calculate the auto-correlation of a given gene in the kNN graph based on the chosen embedding. This can be used to identify genes that have strong local correlations and vary across the kNN graph.

Usage

hotspot_autocor_sc(
  object,
  embd_to_use = "pca",
  use_knn = TRUE,
  hotspot_params = params_sc_hotspot(),
  no_embd_to_use = NULL,
  cells_to_take = NULL,
  genes_to_take = NULL,
  streaming = NULL,
  random_seed = 42L,
  .verbose = TRUE
)

Arguments

object

SingleCells or MetaCells class.

embd_to_use

String. The embedding to use. Defaults to "pca".

use_knn

Boolean. Shall the internal kNN be used. If set to yes, you need to ensure consistency. If you provide cells_to_take, the function will regenerate the kNN graph with these cells.

hotspot_params

List with hotspot parameters, see params_sc_hotspot() with the following elements:

  • model - String. Which of the available models to use for the gene expression. Choices are one of c("danb", "normal", "bernoulli").

  • normalise - Boolean. Shall the data be normalised.

  • weighted_graph - Boolean. Shall the Gaussian kernel be applied to the neighbour distances. If FALSE, every retained edge weighs one.

  • neighborhood_factor - Float. Kernel width for weighted_graph.

  • knn - List of kNN parameters. See params_knn_defaults() for available parameters and their defaults.

no_embd_to_use

Optional integer. Number of embedding dimensions to use. If NULL all will be used.

cells_to_take

Optional string vector. If you want to only use selected cells. If NULL will default to all cells_to_keep in the class.

genes_to_take

Optional string vector. If you wish to limit the search to a subset of genes. If NULL will default to all genes in the class.

streaming

Optional Boolean. Shall the data be streamed in. Useful for larger data sets where you wish to avoid loading in the whole data. If NULL, will automatically detect. Ignored for MetaCells, which are held in memory.

random_seed

Integer. Used for reproducibility.

.verbose

Boolean or integer. Controls verbosity and returns run times. FALSE -> quiet, TRUE or 1L -> normal verbosity, 2L -> detailed verbosity.

Value

A data.table with the auto-correlations on a per gene basis and various statistics.

Details

Should a gene not be found in sufficient cells, the gene will be automatically filtered out from the results. This can occur for example if you have filtered out the cells that contain a given gene. The underlying genes are still available, but the cells that might contain them are not included.

Whether the neighbour distances need squaring before the kernel sees them follows from the metric. With use_knn = TRUE it is taken from the metric stored on the cached kNN graph, otherwise from ann_dist in hotspot_params.

Examples

# local auto-correlation of every gene on the cached kNN graph
sc <- demo_single_cells()
res <- hotspot_autocor_sc(sc, .verbose = FALSE)
head(res, 3)
#>    gene_id  gaerys_c  z_score          pval           fdr
#>     <char>     <num>    <num>         <num>         <num>
#> 1: gene_01 0.4152188 29.01265 2.278132e-185 8.136184e-185
#> 2: gene_02 0.4358798 31.03719 8.494589e-212 5.309118e-211
#> 3: gene_03 0.4703848 32.60140 1.959130e-233 1.632609e-232

unlink(sc@dir_data, recursive = TRUE, force = TRUE)