Simulated Patient Data for Binary Classification

Description

This dataset contains simulated data for a binary classification problem, representing patient data with clinical, proteomics, and RNA variables. The data is organized into three blocks of variables: clinical variables, proteomics variables, and RNA variables. The outcome is a binary variable generated based on a logistic function.

Usage

Pen_Data

Format

A data frame with 406 rows and 325 columns: The dataset includes:

• 5 clinical variables

• 174 proteomic variables

• 145 RNA variables

• Outcome variable Pen_out

Source

Simulated dataset for package examples

Extract coefficients from a priorityelasticnet object

Description

Extract coefficients from a priorityelasticnet object

Usage

## S3 method for class 'priorityelasticnet'
coef(object, ...)

Arguments

Value

List with the coefficients and the intercepts

priorityelasticnet with several block specifications

Description

Runs priorityelasticnet for a list of block specifications and gives the best results in terms of cv error.

Usage

cvm_priorityelasticnet(
  X,
  Y,
  weights,
  family,
  type.measure,
  blocks.list,
  max.coef.list = NULL,
  block1.penalization = TRUE,
  lambda.type = "lambda.min",
  standardize = TRUE,
  nfolds = 10,
  foldid,
  cvoffset = FALSE,
  cvoffsetnfolds = 10,
  alpha = 1,
  ...
)

Arguments

Value

object of class cvm_priorityelasticnet with the following elements. Any elements that are lists contain the detailed results (such as coefficients or cross-validation errors) corresponding to each penalized block of the final, optimal model.

lambda.ind

list with indices of lambda for lambda.type.

lambda.type

type of lambda which is used for the predictions.

lambda.min

list with values of lambda for lambda.type.

min.cvm

list with the mean cross-validated errors for lambda.type.

nzero

list with numbers of non-zero coefficients for lambda.type.

glmnet.fit

list of fitted glmnet objects.

name

a text string indicating type of measure.

block1unpen

if block1.penalization = FALSE, the results of either the fitted glm or coxph object.

best.blocks

character vector with the indices of the best block specification.

best.blocks.indices

list with the indices of the best block specification ordered by best to worst.

best.max.coef

A vector containing the maximum number of non-zero coefficients (selected variables) allowed for each penalized block, corresponding to the model chosen as optimal by cross-validation best.blocks.

best.model

complete priorityelasticnet model of the best solution.

coefficients

coefficients according to the results obtained with best.blocks.

call

the function call.

Note

The function description and the first example are based on the R package ipflasso.

Construct control structures for handling of missing data for priorityelasticnet

Description

Construct control structures for handling of missing data for priorityelasticnet

Usage

missing.control(
  handle.missingdata = c("none", "ignore", "impute.offset"),
  offset.firstblock = c("zero", "intercept"),
  impute.offset.cases = c("complete.cases", "available.cases"),
  nfolds.imputation = 10,
  lambda.imputation = c("lambda.min", "lambda.1se"),
  perc.comp.cases.warning = 0.3,
  threshold.available.cases = 30,
  select.available.cases = c("maximise.blocks", "max")
)

Arguments

Value

list with control parameters

Predictions from priorityelasticnet

Description

Makes predictions for a priorityelasticnet object. It can be chosen between linear predictors or fitted values.

Usage

## S3 method for class 'priorityelasticnet'
predict(
  object,
  newdata = NULL,
  type = c("link", "response"),
  handle.missingtestdata = c("none", "omit.prediction", "set.zero", "impute.block"),
  include.allintercepts = FALSE,
  use.blocks = "all",
  ...
)

Arguments

Details

handle.missingtestdata specifies how to deal with missing data. The default none cannot handle missing data, omit.prediction does not make a prediction for observations with missing values and return NA. set.zero ignores the missing data for the calculation of the prediction (the missing value is set to zero). impute.block uses an imputation model to impute the offset of a missing block. This only works if the priorityelasticnet object was fitted with handle.missingdata = "impute.offset". If impute.offset.cases = "complete.cases" was used, then every observation can have only one missing block. For observations with more than one missing block, NA is returned. If impute.offset.cases = "available.cases" was used, the missingness pattern in the test data has to be the same as in the train data. For observations with an unknown missingness pattern, NA is returned.

Value

Predictions that depend on type.

Examples

pl_bin <- priorityelasticnet(X = matrix(rnorm(50*190),50,190), Y = rbinom(50,1,0.5),
                       family = "binomial", type.measure = "auc",
                       blocks = list(block1=1:13,block2=14:80, block3=81:190),
                       block1.penalization = TRUE, lambda.type = "lambda.min",
                       standardize = FALSE, nfolds = 3, alpha = 1)

newdata_bin <- matrix(rnorm(10*190),10,190)

predict(object = pl_bin, newdata = newdata_bin, type = "response")

Priority Elastic Net for High-Dimensional Data

Description

This function performs penalized regression analysis using the elastic net method, tailored for high-dimensional data with a known group structure. Fits a priority elastic net model for multi-block or multi-omics data. An additional Shiny application for model evaluation and weighted threshold optimization is available through the weightedThreshold function.

Usage

priorityelasticnet(
  X,
  Y,
  weights = NULL,
  family = c("gaussian", "binomial", "cox", "multinomial"),
  alpha = 0.5,
  type.measure,
  blocks,
  max.coef = NULL,
  block1.penalization = TRUE,
  lambda.type = "lambda.min",
  standardize = TRUE,
  nfolds = 10,
  foldid = NULL,
  cvoffset = FALSE,
  cvoffsetnfolds = 10,
  mcontrol = missing.control(),
  scale.y = FALSE,
  return.x = TRUE,
  adaptive = FALSE,
  initial_global_weight = TRUE,
  verbose = FALSE,
  ...
)

Arguments

Value

A list with the following components:

Note

Ensure that glmnet version >= 2.0.13 is installed. The function does not support single missing values within a block.

References

Musib, L., Coletti, R., Lopes, M.B., Mouriño, H. & Carrasquinha, E. (2024). Priority-Elastic net for binary disease outcome prediction based on multi-omics data. BioData Mining, 17(45).https://doi.org/10.1186/s13040-024-00401-0 Klau, S., Jurinovic, V., Hornung, R., Herold, T. & Boulesteix, A.-L. (2018). Priority-Lasso: a simple hierarchical approach to the prediction of clinical outcome using multi-omics data. BMC Bioinformatics, 19, 322. https://doi.org/10.1186/s12859-018-2344-6 Zou, H., & Hastie, T. (2005). Regularization and variable selection via the elastic net. Journal of the Royal Statistical Society: Series B (Statistical Methodology), 67(2), 301–320. https://doi.org/10.1111/j.1467-9868.2005.00527.x

Examples

  # Simulation of multinomial data:
  set.seed(123)
  n <- 100
  p <- 50
  k <- 3
  x <- matrix(rnorm(n * p), n, p)
  y <- sample(1:k, n, replace = TRUE)
  y <- factor(y)
  blocks <- list(bp1 = 1:10, bp2 = 11:30, bp3 = 31:50)
  
  # Run priorityelasticnet:
  fit <- priorityelasticnet(x, y, family = "multinomial", alpha = 0.5, 
                     type.measure = "class", blocks = blocks,
                     block1.penalization = TRUE, lambda.type = "lambda.min", 
                     standardize = TRUE, nfolds = 5, 
                     adaptive = FALSE)
                     
   fit$coefficients

A Shiny App for Model Evaluation and Weighted Threshold Optimization

Description

This function starts a Shiny application that enables users to interactively adjust the threshold for binary classification and view related metrics, the confusion matrix, ROC curve, and PR curve. The app also includes a feature for calculating the optimal threshold using a weighted version of Youden's J-statistic.

Usage

weightedThreshold(object, ...)

Arguments

Details

To calculate the optimal threshold, a weighted version of Youden's J-statistic (Youden, 1950) is used. The optimal cutoff is the threshold that maximizes the distance from the identity (diagonal) line. The function optimizes the metric (w * sensitivity + (1 - w) * specificity), where 'w' is the weight parameter adjusted using the second slider. After selecting the desired value on the optimal threshold slider, the user must press the "Set" button to update the threshold slider with the calculated optimal value. Metrics will then be automatically recalculated based on the user's selection. This function adapted from 'Monahov, A. (2021). Model Evaluation with Weighted Threshold Optimization (and the “mewto” R package). Available at SSRN 3805911.'

#' @references Musib, L., Coletti, R., Lopes, M.B., Mouriño, H. & Carrasquinha, E. (2024). Priority-Elastic net for binary disease outcome prediction based on multi-omics data. BioData Mining, 17(45).https://doi.org/10.1186/s13040-024-00401-0

Value

No return value. This function launches a Shiny application for interactive model evaluation with weighted threshold optimization. It is intended for user interaction rather than returning a computed value. The Shiny app provides an interactive interface to visualize model performance metrics and optimize thresholds for classification models based on user-defined criteria.

Examples

  # Simulation of Pen_data data:
  data("Pen_Data", package = "priorityelasticnet")
  blocks <- list(block1 = 1:5, block2 = 6:179, block3 = 180:324)
  
  # Run priorityelasticnet:
  fit_bin <- priorityelasticnet(X = as.matrix(Pen_Data[, 1:324]), 
                                Y = Pen_Data[, 325],
                                family = "binomial", 
                                alpha = 0.5, 
                                type.measure = "auc",
                                blocks = blocks,
                                standardize = FALSE)
                     
   weightedThreshold(fit_bin)