This function applies the PhenoGraph pipeline given an embedding matrix. The pipeline creates the exact nearest-neighbour graph using the RANN package, then applies the Leiden algorithm to find the clusters. The pipeline can be run multiple times while changing the random seed, to ensure the robustness of the resulting clusters.
Usage
clustering_pipeline(
embedding,
n_neighbours = seq(from = 5, to = 50, by = 5),
graph_type = "snn",
prune_value = -1,
resolutions = list(RBConfigurationVertexPartition = seq(from = 0.1, to = 2, by = 0.1),
RBERVertexPartition = NULL, ModularityVertexPartition = NULL),
number_iterations = 5,
seeds = NULL,
number_repetitions = 100,
merge_identical_partitions = TRUE,
memory_log_file = NULL
)Arguments
- embedding
The embedding matrix where each row represents a point and each column a dimension.
- n_neighbours
A vector of integers representing the number of number of neighbours to be used to build the adjacency graph.
- graph_type
The type of graph. Can be either "nn" - directed unweighted graph or "snn" - undirected weighted graph, where the weights are calculated using the JSI score.
- prune_value
The value used to prune the edges of the graph. If the value is negative, the function will determine the highest pruning value that will keep the graph connected. Defaults to -1.
- resolutions
A list of vectors representing the resolution parameters used in the community detection algorithm. Each name of the list is associated with a quality function. The values of the list are the resolution parameters to be used. The available choices in terms of quality functions are "RBConfigurationVertexPartition", "RBERVertexPartition" and "ModularityVertexPartition".
- number_iterations
How many iterations the Leiden algorithm should run. Defaults to 5.
- seeds
A vector of integers representing the random seeds to be used in the community detection algorithm. If NULL, the function will generate the seeds based on the number of repetitions. Defaults to NULL.
- number_repetitions
The number of repetitions the community detection pipeline should run. Defaults to 100.
- merge_identical_partitions
Logical indicating if the function should merge the partitions that are identical and group them by the number of clusters. Defaults to TRUE.
- memory_log_file
The path to the file where the memory usage of the function should be logged. Defaults to NULL.
Value
A list of lists containing the clusters found for each combination
of number of neighbours and quality functions. If merge_identical_partitions
is set to TRUE, the last level will contain the partitions grouped by the
number of clusters, alongside with their overall Element-Centric Consistency
(ECC) score.