Skip to contents

Identify Independent Features in a Numeric Matrix

tree_and_independent_features.Rd

This function identifies independent features using Spearman's rho correlation distances, and a dendrogram tree cut step.

Usage

tree_and_independent_features(
  data,
  tree_cut_height = 0.5,
  features_exclude = NULL,
  feature_selection = "max_var_exp",
  cores = NULL,
  fast = FALSE
)

Arguments

data

matrix, the 'omics data matrix. samples in row, features in columns

tree_cut_height

the tree cut height. A value of 0.2 (1-Spearman's rho) is equivalent to saying that features with a rho >= 0.8 are NOT independent.

features_exclude

character, vector of feature id indicating features to exclude from the sample and PCA summary analysis but keep in the data

feature_selection

character. Method for selecting a representative feature from each correlated feature cluster.

cores

number of cores available for parallelism; the default null will try find the maximum available cores - 1; set to 1 for linear, but potentially slow, computation of the correlation matrix.

fast

If TRUE, accelerates correlation computation by imputing missing values to the column minimum, pre-ranking all columns, and computing Pearson correlation on ranked data (approximating Spearman). Substantially faster than exact Spearman at large feature dimensions (\(p > 5000\)) but assumes missing data are missing at random. Features with high missingness will have inflated rank ties at the median (ensure these are filtered out appropriately with the missingness option). Default FALSE. One of:

"max_var_exp"

(Default) Selects the feature with the highest sum of absolute Spearman correlations to other features in the cluster; effectively the feature explaining the most shared variance.

"least_missingness"

Selects the feature with the fewest missing values within the cluster.

Value

A list with the following components:

data

A `data.frame` with:

  • `feature_id`: Feature (column) names from the input matrix.

  • `k`: The cluster index assigned to each feature after tree cutting.

  • `independent_features`: Logical indicator of whether the feature was selected as an independent (representative) feature.

tree

A `hclust` object representing the hierarchical clustering of the features based on 1 - |Spearman's rho| distance.