Skip to contents

Impute Missing Exposure and Omics Data in a MultiAssayExperiment

Source: R/run_impute_missing.R
run_impute_missing.Rd

Performs missing data imputation on both exposure variables (from colData) and omics datasets (from experiments) within a MultiAssayExperiment object.

Usage

run_impute_missing(
  expomicset,
  exposure_impute_method = "median",
  exposure_cols = NULL,
  omics_impute_method = NULL,
  omics_to_impute = NULL
)

Arguments

expomicset

A MultiAssayExperiment object containing exposures and omics data.

exposure_impute_method

Character. Imputation method to use for exposure variables. Defaults to "median".

exposure_cols

Character vector. Names of columns in colData to impute. If NULL, all numeric columns are used.

omics_impute_method

Character. Imputation method to use for omics data. Defaults to "knn".

omics_to_impute

Character vector. Names of omics datasets to impute. If NULL, all omics datasets are included.

Value

A MultiAssayExperiment object with imputed exposure and/or omics data.

Details

For exposures, numeric columns in colData are imputed using the selected method. For omics data, assays are selected and imputed individually.

Supported imputation methods include:

  • "median": Median imputation using naniar::impute_median_all

  • "mean": Mean imputation using naniar::impute_mean_all

  • "knn": k-nearest neighbor imputation using impute::impute.knn

  • "mice": Multiple imputation using chained equations (mice::mice)

  • "dep": MinProb imputation for proteomics using DEP::impute

  • "missforest": Random forest-based imputation using missForest::missForest

  • "lod_sqrt2": Substitution of missing values with LOD/sqrt(2), where LOD is the smallest non-zero value per variable

Examples

# Create example data
mae <- make_example_data(
    n_samples = 20,
    return_mae = TRUE
)
#> Ensuring all omics datasets are matrices with column names.
#> Creating SummarizedExperiment objects.
#> Creating MultiAssayExperiment object.
#> MultiAssayExperiment created successfully.

# Introduce some missingness
MultiAssayExperiment::colData(mae)$exposure_pm25[sample(1:20, 5)] <- NA

# Filter features and exposures with high missingness
mae <- run_impute_missing(
    expomicset = mae,
    exposure_impute_method = "median"
)