Skip to contents

Cluster terms in prior knowledge by set overlap

Source: R/RefactorPriorKnoweldge.R
cluster_pk.Rd

Cluster terms in prior knowledge by set overlap

Usage

cluster_pk(
  data,
  metadata_info = c(metabolite_column = "MetaboliteID", pathway_column = "term"),
  similarity = c("jaccard", "overlap_coefficient", "correlation"),
  correlation_method = "pearson",
  input_format = c("long", "enrichment"),
  delimiter = "/",
  threshold = 0.5,
  plot_threshold = 0,
  clust = c("components", "community", "hierarchical"),
  hclust_method = "average",
  min = 2,
  plot_name = "ClusterGraph",
  max_nodes = 10000,
  min_degree = 1,
  node_size_column = NULL,
  show_density = FALSE,
  save_plot = "png",
  print_plot = FALSE,
  path = NULL
)

Arguments

data

Long data frame with one ID per row, or enrichment-style table with a delimited metabolite list per term (see input_format).

metadata_info

List with entries metabolite_column (metabolite ID column or delimited list column) and pathway_column (term column). Defaults to c(metabolite_column = "MetaboliteID", pathway_column = "term").

similarity

Similarity measure between term ID sets. Options: "jaccard" (default), "overlap_coefficient", or "correlation". Jaccard similarity is |A intersect B| / |A union B|. Overlap coefficient is |A intersect B| / min(|A|, |B|). Jaccard is stricter for large sets, while overlap_coefficient is more permissive for nested sets.

correlation_method

Correlation method when similarity = "correlation". One of "pearson", "spearman", "kendall". Ignored otherwise.

input_format

Input format of data. Use "long" for one ID per row (default) or "enrichment" for one term per row with a delimited ID list. The metabolite_column entry in metadata_info is interpreted accordingly.

delimiter

Delimiter for metabolite ID lists when input_format = "enrichment". Ignored for input_format = "long". Default = "/".

threshold

Similarity cutoff for keeping edges (applies to all clustering modes). Default = 0.5.

plot_threshold

Similarity cutoff for plotting edges in viz_graph. Default = 0 (plot all edges with similarity > 0).

clust

Clustering strategy: "components" (connected components on thresholded unweighted graph), "community" (Louvain on thresholded weighted graph), or "hierarchical" (hclust on distance = 1 - similarity).

hclust_method

Linkage method for hierarchical clustering. One of "average" (default), "single", "complete", "ward.D", "ward.D2", "mcquitty", "median", "centroid". Used only when clust = "hierarchical".

min

Minimum cluster size; smaller clusters are relabeled to "None". Default = 2.

plot_name

Optional: String added to output files of the plot. Default = "ClusterGraph".

max_nodes

Optional: Maximum nodes for plotting. If set, keeps nodes from the largest component up to this limit (by degree). Used only for the graph plot. Default = 10000.

min_degree

Optional: Minimum degree filter for graph plotting. Used only for the graph plot. Default = 1.

node_size_column

Optional: Numeric column name from data used to scale node sizes in the graph. Aggregated per term (mean) when multiple rows map to the same term. Default = NULL.

show_density

Optional: If TRUE, add a hull background per cluster to the graph. Default = FALSE.

save_plot

Optional: Select the file type of output plots. Options are svg, pdf, png or NULL. Default = "svg"

print_plot

Optional: If TRUE prints an overview of resulting plots. Default = FALSE

path

Optional: String which is added to the resulting folder name. default: NULL

Value

A list with:

data

Input data with a cluster column added.

cluster_summary

Summary of cluster sizes and percentages.

clusters

Named vector of term -> cluster assignment.

similarity_matrix

Term-by-term similarity matrix.

distance_matrix

Term-by-term distance matrix (1 - similarity).

node_sizes

Named numeric vector of node sizes used in plotting (or NULL).

graph_plot

Graph plot returned by viz_graph.

Examples

# Create toy pathway data in long format
toy_pw <- data.frame(
    MetaboliteID = c("C1", "C2", "C3", "C1", "C2", "C4", "C3", "C4", "C5"),
    term = c("pA", "pA", "pA", "pB", "pB", "pB", "pC", "pC", "pC")
)

r <- cluster_pk(
    toy_pw,
    metadata_info = c(
        metabolite_column = "MetaboliteID",
        pathway_column = "term"
    ),
    input_format = "long",
    similarity = "jaccard",
    threshold = 0.1,
    clust = "community",
    min = 1,
    save_plot = NULL,
    print_plot = FALSE
)