Create Missingness Proportion Matrix

Creates a matrix where each entry represents the proportion of missing values for each sample–feature combination across multiple timepoints. Each sample will have one proportion value per feature. Features may have repeated time points (columns named like feature_1, feature_2, ...). This matrix can be used with cluster_on_missing_prop() to group samples with similar missingness patterns.

Usage

create_missingness_prop_matrix(
  data,
  index_col = NULL,
  cols_ignore = NULL,
  na_values = c(NA, NaN, Inf, -Inf),
  repeat_feature_names = character(0)
)

Arguments

data

Data frame or matrix containing the input data with potential missing values.

index_col

Character scalar. Name of an index column to exclude from analysis (optional). If supplied and present, it will be removed from analysis; row names are preserved as-is.

cols_ignore

Character vector of column names to exclude from the proportion matrix (optional).

na_values

Vector of values to treat as missing in addition to standard missing values. Defaults to c(NA, NaN, Inf, -Inf).

repeat_feature_names

Character vector of "base" feature names that have repeated timepoints. Repeat measurements must be in the form <feature>_<timepoint> where <feature> is alphanumeric (and may include dots) and <timepoint> is an integer (e.g., "CRP_1").

Value

A numeric matrix of dimension nrow(data) by n_features, where rows are samples and columns are features (base names). Entries are per-sample missingness proportions in [0, 1]. The returned matrix has an attribute "feature_columns_map": a named list mapping each output feature to the source columns used to compute its proportion.

Examples

df <- data.frame(
  id = paste0("s", 1:4),
  CRP_1 = c(1.2, NA, 2.1, NaN),
  CRP_2 = c(NA, NA, 2.0, 1.9),
  IL6_1 = c(0.5, 0.7, Inf, 0.4),
  IL6_2 = c(0.6, -Inf, 0.8, 0.5),
  Albumin = c(3.9, 4.1, 4.0, NA)
)

m <- create_missingness_prop_matrix(
  data = df,
  index_col = "id",
  cols_ignore = NULL,
  repeat_feature_names = c("CRP", "IL6")
)

dim(m)         # 4 x 3 (CRP, IL6, Albumin)
#> [1] 4 3
m[ , "CRP"]    # per-sample proportion missing across CRP_1 and CRP_2
#>   1   2   3   4 
#> 0.5 1.0 0.0 0.5 
attr(m, "feature_columns_map")
#> NULL