imp4p (version 1.2)

impute.mix: Imputation using a decision rule under an assumption of a mixture of MCAR and MNAR values.

Description

This function allows imputing data sets with a MCAR-devoted algorithm and a MNAR-devoted algorithm using probabilities that missing values are MCAR. If such a probability is superior to a chosen threshold, then the MCAR-devoted algorithm is used, otherwise it is the MNAR-devoted algorithm. For details, see Giai Gianetto, Q. et al. (2020) (doi: 10.1101/2020.05.29.122770).

Usage

impute.mix(tab, prob.MCAR, threshold, conditions, repbio=NULL, reptech=NULL,
methodMCAR="mle",nknn=15,weight=1, selec="all", ind.comp=1, progress.bar=TRUE, q=0.95,
ncp.max=5, maxiter = 10, ntree = 100, variablewise = FALSE, decreasing = FALSE,
verbose = FALSE, mtry = floor(sqrt(ncol(tab))), replace = TRUE,classwt = NULL,
cutoff = NULL, strata = NULL, sampsize = NULL, nodesize = NULL, maxnodes = NULL,
xtrue = NA, parallelize = c('no', 'variables', 'forests'),
methodMNAR="igcda", q.min = 0.025, q.norm = 3, eps = 0, distribution = "unif",
param1 = 3, param2 = 1, R.q.min=1)

Arguments

tab

A data matrix containing numeric and missing values. Each column of this matrix is assumed to correspond to an experimental sample, and each row to an identified peptide.

prob.MCAR

A matrix of probabilities that each missing value is MCAR. For instance such a matrix can be obtained from the function prob.mcar.tab of this package.

threshold

A value such that if the probability that a missing value is MCAR is superior to it, then a MCAR-devoted algorithm is used, otherwise it is a MNAR-devoted algorithm that is used.

conditions

A vector of factors indicating the biological condition to which each column (experimental sample) belongs.

repbio

A vector of factors indicating the biological replicate to which each column belongs. Default is NULL (no experimental design is considered).

reptech

A vector of factors indicating the technical replicate to which each column belongs. Default is NULL (no experimental design is considered).

methodMCAR

The method used for imputing MCAR data. If methodi="mle" (default), then the impute.mle function is used (imputation using an EM algorithm). If methodi="pca", then the impute.PCA function is used (imputation using Principal Component Analysis). If methodi="rf", then the impute.RF function is used (imputation using Random Forest). Else, the impute.slsa function is used (imputation using Least Squares on nearest neighbours).

methodMNAR

The method used for imputing MNAR data. If methodMNAR="igcda" (default), then the impute.igcda function is used. Else, the impute.pa function is used.

nknn

The number of nearest neighbours used in the SLSA algorithm (see impute.slsa).

weight

The way of weighting in the algorithm (see impute.slsa).

selec

A parameter to select a part of the dataset to find nearest neighbours between rows. This can be useful for big data sets (see impute.slsa).

ind.comp

If ind.comp=1, only nearest neighbours without missing values are selected to fit linear models (see impute.slsa). Else, they can contain missing values.

progress.bar

If TRUE, a progress bar is displayed.

q

A quantile value (see impute.igcda).

ncp.max

parameter of the impute.PCA function.

maxiter

parameter of the impute.RF function.

ntree

parameter of the impute.RF function.

variablewise

parameter of the impute.RF function.

decreasing

parameter of the impute.RF function.

verbose

parameter of the impute.RF function.

mtry

parameter of the impute.RF function.

replace

parameter of the impute.RF function.

classwt

parameter of the impute.RF function.

cutoff

parameter of the impute.RF function.

strata

parameter of the impute.RF function.

sampsize

parameter of the impute.RF function.

nodesize

parameter of the impute.RF function.

maxnodes

parameter of the impute.RF function.

xtrue

parameter of the impute.RF function.

parallelize

parameter of the impute.RF function.

q.min

parameter of the impute.pa function.

q.norm

parameter of the impute.pa function.

eps

parameter of the impute.pa function.

distribution

parameter of the impute.pa function.

param1

parameter of the impute.pa function.

param2

parameter of the impute.pa function.

R.q.min

parameter of the impute.pa function.

Value

The input matrix tab with imputed values instead of missing values.

Details

The missing values for which prob.MCAR is superior to a chosen threshold are imputed with one of the MCAR-devoted imputation methods (impute.mle, impute.RF, impute.PCA or impute.slsa). The other missing values are considered MNAR and imputed with impute.igcda. More details and explanations can be bound in Giai Gianetto (2020).

References

Giai Gianetto, Q., Wieczorek S., Cout<U+00E9> Y., Burger, T. (2020). A peptide-level multiple imputation strategy accounting for the different natures of missing values in proteomics data. bioRxiv 2020.05.29.122770; doi: 10.1101/2020.05.29.122770

Examples

# NOT RUN {
#Simulating data
res.sim=sim.data(nb.pept=2000,nb.miss=600);

#Fast imputation of missing values with the impute.rand algorithm
dat.rand=impute.rand(tab=res.sim$dat.obs,conditions=res.sim$condition);

#Estimation of the mixture model
res=estim.mix(tab=res.sim$dat.obs, tab.imp=dat.rand, conditions=res.sim$condition);

#Computing probabilities to be MCAR
born=estim.bound(tab=res.sim$dat.obs,conditions=res.sim$condition);
proba=prob.mcar.tab(born$tab.upper,res);

#Imputation under the assumption of MCAR and MNAR values
tabi=impute.mix(tab=res.sim$dat.obs, prob.MCAR=proba, threshold=0.5, conditions=res.sim$conditions,
repbio=res.sim$repbio, methodMCAR="slsa", methodMNAR="igcda", nknn=15, weight=1, selec="all",
ind.comp=1, progress.bar=TRUE);
# }