imputeAssay: Impute missing values in a 'matrix'

In tnaake/MatrixQCvis: Shiny-based interactive data-quality exploration for omics data

Impute missing values in a matrix

Description

The function impute imputes missing values based on one of the following principles: Bayesian missing value imputation (BPCA), k-nearest neighbor averaging (kNN), Malimum likelihood-based imputation method using the EM algorithm (MLE), replacement by the smallest non-missing value in the data (Min), replacement by the minimal value observed as the q-th quantile (MinDet, default q = 0.01), and replacement by random draws from a Gaussian distribution centred to a minimal value (MinProb).

Usage

imputeAssay(
  a,
  method = c("BPCA", "kNN", "MLE", "Min", "MinDet", "MinProb", "none")
)

Arguments

a	matrix with samples in columns and features in rows
method	character, one of "BPCA", "kNN", "MLE", "Min", "MinDet", "MinProb", or "none"

Details

BPCA wrapper for pcaMethods::pca with methods = "bpca". BPCA is a missing at random (MAR) imputation method.

kNN wrapper for impute::impute.knn with k = 10, rowmax = 0.5, colmax = 0.5, maxp = 1500. kNN is a MAR imputation method.

MLE wrapper for imputeLCMD::impute.MAR with method = "MLE", model.selector = 1/imputeLCMD::impute.wrapper.MLE. MLE is a MAR imputation method.

Min imputes the missing values by the observed minimal value of x. Min is a missing not at random (MNAR) imputation method.

MinDet is a wrapper for imputeLCMD::impute.MinDet with q = 0.01. MinDet performs the imputation using a deterministic minimal value approach. The missing entries are replaced with a minimal value, estimated from the q-th quantile from each sample. MinDet is a MNAR imputation method.

MinProb is a wrapper for imputeLCMD::impute.MinProb with q = 0.01 and tune.sigma = 1. MinProb performs the imputation based on random draws from a Gaussion distribution with the mean set to the minimal value of a sample. MinProb is a MNAR imputation method.

MinProb does not impute values (not available within shiny application).

Value

matrix

Examples

a <- matrix(seq_len(100), nrow = 10, ncol = 10, 
    dimnames = list(seq_len(10), paste("sample", seq_len(10))))
a[c(1, 5, 8), seq_len(5)] <- NA

imputeAssay(a, method = "kNN")
imputeAssay(a, method = "Min")
imputeAssay(a, method = "MinDet")
imputeAssay(a, method = "MinProb")

tnaake/MatrixQCvis documentation built on Nov. 4, 2024, 7:42 a.m.