Type: Package
Title: Methods and Reproducible Workflows for Partial Least Squares with Missing Data
Version: 0.2.0
Date: 2026-04-07
Depends: R (≥ 4.1.0)
Imports: mice, plsRglm, stats, utils, VIM
Suggests: bcv, knitr, mlbench, plsdof, rmarkdown, testthat (≥ 3.0.0)
Author: Titin Agustin Nengsih [aut], Frederic Bertrand [aut, cre], Myriam Maumy-Bertrand [aut]
Maintainer: Frederic Bertrand <frederic.bertrand@lecnam.net>
Description: Methods-first tooling for reproducing and extending the partial least squares regression studies on incomplete data described in Nengsih et al. (2019) <doi:10.1515/sagmb-2018-0059>. The package provides simulation helpers, missingness generators, imputation wrappers, component-selection utilities, real-data diagnostics, and reproducible study orchestration for Nonlinear Iterative Partial Least Squares (NIPALS)-Partial Least Squares (PLS) workflows.
License: GPL-3
Encoding: UTF-8
LazyData: true
VignetteBuilder: knitr
Config/testthat/edition: 3
RoxygenNote: 7.3.3
URL: https://fbertran.github.io/missPLS/, https://github.com/fbertran/missPLS
BugReports: https://github.com/fbertran/missPLS/issues
NeedsCompilation: no
Packaged: 2026-04-08 09:17:07 UTC; bertran7
Repository: CRAN
Date/Publication: 2026-04-13 15:00:02 UTC

missPLS: Methods and Reproducible Workflows for Partial Least Squares with Missing Data

Description

Methods-first tooling for reproducing and extending the partial least squares regression studies on incomplete data described in Nengsih et al. (2019) doi:10.1515/sagmb-2018-0059. The package provides simulation helpers, missingness generators, imputation wrappers, component-selection utilities, real-data diagnostics, and reproducible study orchestration for Nonlinear Iterative Partial Least Squares (NIPALS)-Partial Least Squares (PLS) workflows.

Author(s)

Maintainer: Frederic Bertrand frederic.bertrand@lecnam.net

Authors:

See Also

Useful links:

Add missing values to a predictor matrix

Description

Create MCAR or MAR missingness on the predictor matrix x. Missingness is generated column-wise so that each predictor receives approximately the same missing-data proportion, matching the simulation strategy used in the original work.

Usage

add_missingness(
  x,
  y,
  mechanism = c("MCAR", "MAR"),
  missing_prop,
  seed = NULL,
  mar_y_bias = 0.8
)

Arguments

x

Predictor matrix or data frame.

y

Numeric response vector.

mechanism

Missingness mechanism: "MCAR" or "MAR".

missing_prop

Missing-data proportion as a fraction (0.05) or a percentage (5).

seed

Optional random seed. If supplied, it is used only for this call.

mar_y_bias

Proportion of missing values assigned to the upper half of the observed y values under the MAR mechanism.

Value

A list with components x_incomplete, missing_mask, missing_prop, mechanism, and seed.

Examples

sim <- simulate_pls_data(n = 20, p = 10, true_ncomp = 2, seed = 1)
miss <- add_missingness(sim$x, sim$y, mechanism = "MCAR", missing_prop = 10, seed = 2)
mean(is.na(miss$x_incomplete))

Bromhexine dataset

Description

Bromhexine in pharmaceutical syrup used in the article and thesis.

Usage

bromhexine

Format

A misspls_dataset list with components:

name

Dataset name.

x

A numeric ⁠23 x 64⁠ predictor matrix.

y

A numeric response vector of length 23.

data

A data frame with response y and predictors x1 to x64.

source

A short source reference.

preprocessing

Dataset preprocessing notes.

notes

Additional study notes.

Source

Goicoechea and Olivieri (1999a), calibration and test files bundled in extra_docs/pls_data.

Diagnose a real dataset

Description

Compute correlation summaries and VIF-style diagnostics for a packaged real dataset.

Usage

diagnose_real_data(dataset, cor_threshold = 0.7)

Arguments

dataset

A packaged dataset name or misspls_dataset object.

cor_threshold

Absolute-correlation threshold used when reporting predictor pairs and predictor-response associations.

Value

A list with correlation and VIF summaries.

Examples

diag_bromhexine <- diagnose_real_data("bromhexine")
names(diag_bromhexine)

Impute a predictor matrix

Description

Apply one of the imputation strategies used in the article and thesis.

Usage

impute_pls_data(
  x,
  method = c("mice", "knn", "svd"),
  seed = NULL,
  m,
  k = 15L,
  svd_rank = 10L,
  svd_maxiter = 1000L
)

Arguments

x

Incomplete predictor matrix or data frame.

method

Imputation method: "mice", "knn", or "svd".

seed

Optional random seed forwarded to stochastic imputers when supported.

m

Number of imputations for method = "mice". By default this is set to the missing-data percentage rounded to the nearest integer.

k

Number of neighbours for method = "knn".

svd_rank

Target rank for method = "svd".

svd_maxiter

Maximum number of iterations for the fallback SVD routine.

Value

A misspls_imputation object.

Examples

sim <- simulate_pls_data(n = 20, p = 10, true_ncomp = 2, seed = 1)
miss <- add_missingness(sim$x, sim$y, mechanism = "MCAR", missing_prop = 10, seed = 2)
imp <- impute_pls_data(miss$x_incomplete, method = "knn", seed = 3)
length(imp$datasets)

Octane dataset

Description

Octane in gasoline from NIR data used in the article and thesis.

Usage

octane

Format

A misspls_dataset list with components:

name

Dataset name.

x

A numeric ⁠68 x 493⁠ predictor matrix.

y

A numeric response vector of length 68.

data

A data frame with response y and predictors x1 to x493.

source

A short source reference.

preprocessing

Dataset preprocessing notes.

notes

Additional study notes.

Source

Goicoechea and Olivieri (2003), calibration and test files bundled in extra_docs/pls_data.

Complete-case ozone dataset

Description

Los Angeles ozone pollution complete-case dataset used in the article and thesis.

Usage

ozone_complete

Format

A misspls_dataset list with components:

name

Dataset name.

x

A numeric ⁠203 x 12⁠ predictor matrix.

y

A numeric response vector of length 203.

data

A data frame with response y and predictors x1 to x12.

source

A short source reference.

preprocessing

Dataset preprocessing notes.

notes

Additional study notes.

Source

mlbench::Ozone, restricted to the 203 complete observations used in the published analysis.

Run a real-data study

Description

Run a real-data study

Usage

run_real_data_study(
  dataset,
  seed = NULL,
  missing_props = seq(5, 50, 5),
  mechanisms = c("MCAR", "MAR"),
  reps = 1L,
  baseline_reps = 100L,
  max_ncomp = 12L,
  criteria = c("Q2-LOO", "Q2-10fold", "AIC", "AIC-DoF", "BIC", "BIC-DoF"),
  incomplete_methods = c("nipals_standard", "nipals_adaptative"),
  imputation_methods = c("mice", "knn", "svd"),
  folds = 10L,
  mar_y_bias = 0.8
)

Arguments

dataset

A packaged dataset name or misspls_dataset object.

seed

Optional base random seed.

missing_props

Missing-data proportions as fractions or percentages.

mechanisms

Missing-data mechanisms.

reps

Number of replicate missingness draws for each mechanism and proportion.

baseline_reps

Number of repeated complete-data ⁠Q2-10fold⁠ selections used to determine ⁠t**⁠.

max_ncomp

Maximum number of extracted components.

criteria

Criteria evaluated on incomplete and imputed data.

incomplete_methods

Incomplete-data NIPALS workflows.

imputation_methods

Imputation methods.

folds

Number of folds used by "Q2-10fold".

mar_y_bias

MAR bias parameter passed to add_missingness().

Value

A data frame with one row per study run.

Run a simulation study

Description

Run the simulation workflows used in the article and thesis.

Usage

run_simulation_study(
  dimensions = list(c(500L, 100L), c(500L, 20L), c(100L, 20L), c(80L, 25L), c(60L, 33L),
    c(40L, 50L), c(20L, 100L)),
  true_ncomp = c(2L, 4L, 6L),
  missing_props = seq(5, 50, 5),
  mechanisms = c("MCAR", "MAR"),
  reps = 1L,
  seed = NULL,
  max_ncomp = 8L,
  criteria = c("Q2-LOO", "Q2-10fold", "AIC", "AIC-DoF", "BIC", "BIC-DoF"),
  incomplete_methods = c("nipals_standard", "nipals_adaptative"),
  imputation_methods = c("mice", "knn", "svd"),
  folds = 10L,
  mar_y_bias = 0.8
)

Arguments

dimensions

List of ⁠(n, p)⁠ integer pairs.

true_ncomp

Vector of true component counts.

missing_props

Missing-data proportions as fractions or percentages.

mechanisms

Missing-data mechanisms.

reps

Number of replicates.

seed

Optional base random seed.

max_ncomp

Maximum number of extracted components.

criteria

Criteria evaluated on complete and imputed data.

incomplete_methods

Incomplete-data NIPALS workflows.

imputation_methods

Imputation methods.

folds

Number of folds used by "Q2-10fold".

mar_y_bias

MAR bias parameter passed to add_missingness().

Value

A data frame with one row per study run.

Select the number of PLS components

Description

Select the number of components for complete, imputed, or incomplete-data PLS workflows.

Usage

select_ncomp(
  x,
  y,
  method = c("complete", "nipals_standard", "nipals_adaptative"),
  criterion = c("Q2-LOO", "Q2-10fold", "AIC", "AIC-DoF", "BIC", "BIC-DoF"),
  max_ncomp,
  seed = NULL,
  folds = 10L,
  threshold = 0.0975
)

Arguments

x

Predictor matrix, dataset object, or misspls_imputation object.

y

Numeric response vector. This may be omitted when x already contains a response.

method

Selection workflow: "complete", "nipals_standard", or "nipals_adaptative".

criterion

Selection criterion: "Q2-LOO", "Q2-10fold", "AIC", "AIC-DoF", "BIC", or "BIC-DoF".

max_ncomp

Maximum number of components to consider.

seed

Optional random seed used by the cross-validation and imputation aggregation steps.

folds

Number of cross-validation folds used by "Q2-10fold".

threshold

Threshold applied to Q2 criteria.

Value

A one-row data frame describing the selected component count.

Examples

sim <- simulate_pls_data(n = 25, p = 10, true_ncomp = 2, seed = 1)
select_ncomp(sim$x, sim$y, method = "complete", criterion = "AIC", max_ncomp = 4, seed = 2)

Simulate PLS data

Description

Simulate a univariate-response PLS dataset using the Li et al.-style generator available in plsRglm.

Usage

simulate_pls_data(n, p, true_ncomp, seed = NULL, model = "li2002")

Arguments

n

Number of observations.

p

Number of predictors.

true_ncomp

True number of latent components.

seed

Optional random seed. If supplied, it is used only for this call.

model

Simulation model. Only "li2002" is currently supported.

Value

A list with components x, y, data, true_ncomp, seed, and model.

Examples

sim <- simulate_pls_data(n = 20, p = 10, true_ncomp = 2, seed = 42)
str(sim)

Summarize simulation or real-data study results

Description

Summarize simulation or real-data study results

Usage

summarize_simulation_study(results)

Arguments

results

A results data frame returned by run_simulation_study() or run_real_data_study().

Value

A grouped summary data frame.

Examples

sim_results <- run_simulation_study(
  dimensions = list(c(30, 12)),
  true_ncomp = 2,
  missing_props = numeric(0),
  mechanisms = character(0),
  reps = 2,
  seed = 1
)
summarize_simulation_study(sim_results)

Tetracycline dataset

Description

Tetracycline in serum used in the article and thesis.

Usage

tetracycline

Format

A misspls_dataset list with components:

name

Dataset name.

x

A numeric ⁠107 x 101⁠ predictor matrix.

y

A numeric response vector of length 107.

data

A data frame with response y and predictors x1 to x101.

source

A short source reference.

preprocessing

Dataset preprocessing notes.

notes

Additional study notes.

Source

Goicoechea and Olivieri (1999b), calibration and test files bundled in extra_docs/pls_data.