Skip to contents

Combined multiDEGGs filter

Source: R/feature_selection_for_ML.R
multiDEGGs_combined_filter.Rd

This function can be passed to the modifyX parameter of nestcv.train or nestcv.glmnet to use one of the available statistical filters (t-test, wilcoxon, etc.) in combination with multiDEGGs. Single predictors will be selected by the selected statistical filter an paired predictors will be added by multiDEGGs.

Usage

multiDEGGs_combined_filter(
  y,
  x,
  filter_method = "ttest",
  nfilter,
  dynamic_nfilter = TRUE,
  keep_single_genes = FALSE,
  ...
)

Arguments

y

Numeric vector or factor. Response variable (outcome), i.e. the 'metadata' named vector, as passed by nestcv.train or nestcv.glmnet.

x

Predictor variables, i.e. the assayData matrix with genes in columns and IDs in rows, as passed by nestcv.train or nestcv.glmnet.

filter_method

Character string. Statistical filtering method to be used in combination with multiDEGGs for sigle feature selection. Options are: "ttest", "wilcoxon", "ranger", "glmnet", "pls".

nfilter

Integer. Maximum number of features to select.

dynamic_nfilter

Logical. If TRUE nfilter will limit the number of features selected by the statistical filter and the feature space will be augmented by adding ALL the paired predictors found by multiDEGGs. If FALSE nfilter will limit the total number of predictors, with approximately half allocated to pairs and half to single genes.

keep_single_genes

Logical. When dynamic_nfilter = TRUE, determines whether to include single genes selected by multiDEGGs (i.e. the single variables included in the differential pairs) in addition to those from the statistical filter. Default is FALSE.

...

Additional arguments passed to the filtering functions.

Value

An object of class "multiDEGGs_filter" containing:

keep

Character vector of selected single gene names

pairs

Data frame of selected gene pairs with interaction information

Details

The function operates in two modes:

Dynamic Filtering (dynamic_nfilter = TRUE):

  • Selects nfilter single genes using the specified statistical method

  • Finds all significant gene pairs using multiDEGGs

  • Total predictors = nfilter single genes + number of significant pairs

    • If keep_single_genes = TRUE, also includes single genes obtained from pairs found by multiDEGGs

Balanced Selection (dynamic_nfilter = FALSE):

  • Allocates approximately half of nfilter to gene pairs

  • Remaining slots filled with single genes from the statistical filter

  • If fewer pairs are found than allocated, compensates by selecting more single genes

  • Ensures consistent total number of predictors across outer folds

The statistical filtering methods include:

  • "ttest": Two-sample t-test for differential expression

  • "wilcoxon": Wilcoxon rank-sum test

  • "ranger": Random Forest variable importance

  • "glmnet": Elastic net regularization

  • "pls": Partial Least Squares variable importance

Examples

library(nestedcv)
data("synthetic_metadata")
data("synthetic_rnaseqData")

# fit a regularized linear model
# note that nfilter, n_outer_folds, n_inner_folds are set low to keep the
# example lightweight. Adjust these values as needed for your use case.
if (FALSE) { # \dontrun{
fit.glmnet <- nestedcv::nestcv.glmnet(
y = as.numeric(synthetic_metadata$response),
x =  t(synthetic_rnaseqData),
modifyX = "multiDEGGs_combined_filter",
modifyX_options = list(filter_method = "ttest", 
                       nfilter = 5,
                       dynamic_nfilter = TRUE, 
                       keep_single_genes = FALSE),
modifyX_useY = TRUE,
n_outer_folds = 4,
n_inner_folds = 4)

summary(fit.glmnet)
} # }

# fit a random forest model
# NOTE: nfilter, n_outer_folds, n_inner_folds are set low to keep the
# example lightweight. Adjust these values as needed for your use case.
fit.rf <- nestedcv::nestcv.train(
  y = synthetic_metadata$response,
  x = t(synthetic_rnaseqData),
  method = "rf",
  modifyX = "multiDEGGs_combined_filter",
  modifyX_options = list(filter_method = "ttest", 
                         nfilter = 5,
                         dynamic_nfilter = TRUE, 
                         keep_single_genes = FALSE),
  modifyX_useY = TRUE,
  n_outer_folds = 2,
  n_inner_folds = 2
)
#> Fitting final model using CV on whole data
#> Loading required package: ggplot2
#> Loading required package: lattice
#> Duration: 0.8662236 secs

fit.rf$summary
#>                Reference
#> Predicted       Non_responder Responder
#>   Non_responder            57         2
#>   Responder                 1        40
#> 
#>               AUC            Accuracy   Balanced accuracy   
#>            0.9963              0.9700              0.9676