diceR (version 3.0.0)

impute_missing: Impute missing values

Description

Impute missing values from bootstrapped subsampling

Usage

impute_missing(E, data, nk)

Value

If flattened matrix consists of more than one repetition, i.e. it isn't a column vector, then the function returns a matrix of clusterings with complete cases imputed using majority voting, and relabelled, for chosen k.

Arguments

E

4D array of clusterings from consensus_cluster. The number of rows is equal to the number of cases to be clustered, number of columns is equal to the clusterings obtained by different resamplings of the data, the third dimension are the different algorithms and the fourth dimension are cluster sizes.

data

data matrix with samples as rows and genes/features as columns

nk

cluster size to extract data for (single value)

Author

Aline Talhouk

Details

The default output from consensus_cluster will undoubtedly contain NA entries because each replicate chooses a random subset (with replacement) of all samples. Missing values should first be imputed using impute_knn(). Not all missing values are guaranteed to be imputed by KNN. See class::knn() for details. Thus, any remaining missing values are imputed using majority voting.

See Also

Other imputation functions: impute_knn()

Examples

if (FALSE) { # rlang::is_installed("kernlab")
data(hgsc)
dat <- hgsc[1:100, 1:50]
E <- consensus_cluster(dat, nk = 3:4, reps = 10, algorithms = c("hc", "km",
"sc"), progress = FALSE)
sum(is.na(E))
E_imputed <- impute_missing(E, dat, 4)
sum(is.na(E_imputed))
}