Skip to contents

pipeML

A robust R machine learning pipeline for classification tasks and survival analysis

Installation

You can install the development version of pipeML from GitHub with:

# install.packages("pak")
pak::pkg_install("VeraPancaldiLab/pipeML")

Description

pipeML is a robust R-based pipeline for building, evaluating, and interpreting machine learning models in classification tasks. It is designed for fast, user-friendly deployment while maintaining the flexibility and rigor required for research-grade analyses. The pipeline integrates all essential steps — from preprocessing to feature selection, cross-validation, hyperparameter tuning, and model interpretation — into a single, consistent framework (Figure 1).

Figure 1. General structure of the pipeML machine learning pipeline.

Key Features

  • Stratified data partitioning to preserve class balance in both training and testing sets.

  • Iterative Boruta algorithm for robust feature selection.

  • Customizable cross-validation with repeated k-fold CV or stratified

  • Hyperparameter tuning driven by AUROC, AUPRC, or Accuracy.

  • Parallelization support for faster cross-validation and model training across multiple cores.

  • Custom fold construction functions: users can inject their own fold-building logic.

These functions can also accept a bestune argument, which is automatically passed after hyperparameter optimization, allowing seamless retraining on the full training set with the best parameters.

  • Preprocessing utilities for feature filtering (e.g., correlation pruning).

  • Model interpretation via SHAP values for feature importance.

  • Model stacking based on GLM for ensemble learning.

  • Visualization functions for ROC and PR curves, plus performance summaries.

  • Support for 13 machine learning methods, including:

    • Bagged CART
    • Random Forest (RF)
    • C50
    • Logistic regression (LG)
    • CART
    • Naive Bayes (NB)
    • Regularized Lasso
    • Ridge regression
    • Linear Discriminant Analysis (LDA)
    • Regularized Logistic Regression (Elastic net)
    • K-nearest neighbors (KNN)
    • Support vector machine with radial kernel (SVMr)
    • Support vector machine with linear kernel (SVMl)
    • Extreme Gradient Boosting (XGboost)

General usage

These are basic examples which shows you how to use pipeML for different tasks. For a detailed tutorial, see Get started

compute_features.training.ML(): This function is designed for training machine learning models on a single dataset using repeated k-fold cross-validation. It supports feature selection via Boruta, optional model stacking, and flexible hyperparameter tuning and the construction of k-folds stratified by cohorts when this information is available. It can be used when the user do not account with a prediction dataset, in order to train different folds on the same dataset and evaluate performance.

res_ml = compute_features.training.ML(features_train, clinical$Response, "CR", 
                                      metric = "AUROC", stack = F, k_folds = 5, 
                                      n_rep = 10, file_name = "Test", ncores = 2, return = T)

After training, predictions on new data can be computed using the compute_prediction() function. You can specify which metric to maximize when determining the optimal classification threshold. Supported values for maximize include: “Accuracy”, “Precision”, “Recall”, “Specificity”, “Sensitivity”, “F1”, and “MCC”.

pred = compute_prediction(res_ml, features_test, traitData_test$Response, 
                          "CR", stack = F, file.name = "Test", 
                          maximize = "Accuracy", return = T)

compute_features.ML(): This function is intended for training on a dataset and evaluating on a separate test dataset when is available. It automatically computes the prediction using the trained model in the testing set provided. It includes both previous functions.

res = compute_features.ML(tme_features_train[[i]], tme_features_test[[i]], 
                          clinical = traitData, trait = "Response", 
                          trait.positive = "R", metric = "AUROC", stack = F, 
                          k_folds = 2, n_rep = 1, LODO = T, batch_id = "Cohort", 
                          ncores = 2, maximize = "Accuracy", return = F)

Issues

If you encounter any problems or have questions about the package, we encourage you to open an issue here. We’ll do our best to assist you!

Authors

pipeML was developed by Marcelo Hurtado in supervision of Vera Pancaldi and is part of the Pancaldi team. Currently, Marcelo is the primary maintainer of this package.

Citing pipeML

If you use pipeML in a scientific publication, we would appreciate citation to the :

Hurtado M, Pancaldi V (2025). pipeML: A robust R machine learning pipeline for classification tasks and survival analysis. R package version 0.0.1, https://github.com/VeraPancaldiLab/pipeML.