Skip to contents

Identify correlation-based gene modules via graphs

cor_module_graph_final_modules.Rd

This function leverages graph-based clustering to identify gene co-expression modules. The class has the option to sub-cluster large communities within their respective sub graphs, akin to the approach taken by Barrio-Hernandez, et al.

Usage

cor_module_graph_final_modules(
  object,
  resolution = NULL,
  min_size = 10L,
  max_size = 500L,
  subclustering = TRUE,
  random_seed = 123L,
  .graph_params = params_cor_graph(),
  .max_iters = 100L,
  .verbose = TRUE
)

Arguments

object

The class, see BulkCoExp().

resolution

The Leiden resolution parameter you wish to use. If NULL, it will use the optimal one identified by cor_module_graph_check_res(). If nothing can be found, will default to 1.

min_size

Integer. Minimum size of the communities.

max_size

Integer. Maximum size of the communities.

subclustering

Boolean. Shall after a first clustering communities that are too large be further sub clustered. Defaults to TRUE.

random_seed

Integer. Random seed.

.graph_params

List. Parameters for the generation of the (differential) correlation graph, see params_cor_graph(). Contains:

  • Epsilon - Defines the epsilon parameter for the radial basis function. Defaults to 2, but should be ideally optimised.

  • min_cor - Float. Minimum absolute correlation that needs to be observed in either data set. Only relevant for differential correlation-based graphs.

  • fdr_threshold - Float. Maximum FDR for the differential correlation p-value.

  • verbose - Boolean. Controls verbosity of the graph generation.

This parameter is only relevant if you did not run cor_module_graph_check_res().

.max_iters

Integer. If sub clustering is set to TRUE, what shall be the maximum number of iterations. Defaults to 100L.

.verbose

Boolean. Controls the verbosity of the function.

Value

The class with added data to the properties.

References

Barrio-Hernandez, et al., Nat Genet, 2023.

Examples

# finalise the communities at the resolution picked by the scan
mat <- t(synthetic_signal_matrix()$mat)
obj <- BulkCoExp(mat, data.table::data.table(sample_id = rownames(mat)))
obj <- preprocess_bulk_coexp(obj, hvg = 0.3, .verbose = FALSE)
obj <- cor_module_processing(obj, cor_method = "spearman", .verbose = FALSE)
obj <- cor_module_graph_check_res(
  obj,
  resolution_params = params_graph_resolution(number_res = 5L),
  parallel = FALSE,
  .verbose = FALSE
)
obj <- cor_module_graph_final_modules(obj, .verbose = FALSE)
head(get_modules(get_results(obj)))
#>       gene module_id
#>     <char>    <char>
#> 1:  gene60 cluster_1
#> 2: gene139 cluster_2
#> 3: gene196 cluster_2
#> 4: gene176 cluster_2
#> 5: gene203 cluster_3
#> 6:  gene56 cluster_1