inestedcv

inestedcv implements nested cross-validation workflows to provide unbiased model evaluation, hyperparameter tuning, and embedded feature selection for high-dimensional biomedical datasets such as transcriptomics.

Key Features:

• Nested Cross-Validation: Implements a fully nested k × l-fold cross-validation framework with inner loops for hyperparameter tuning and outer loops for unbiased performance assessment to prevent information leakage.
• Embedded Feature Selection: Integrates fast filter functions for feature selection that are nested within the outer CV loop to avoid biasing test-set performance.
• Support for Multiple Models: Uses glmnet to fit lasso and elastic-net regularized linear models and integrates with the caret framework to support additional machine learning algorithms.
• Bayesian Regression Models: Provides Bayesian linear and logistic regression using the horseshoe prior to encourage sparsity in high-dimensional settings.
• High-dimensional Data Handling: Targets P ≫ n scenarios common in transcriptomics and other biomedical research where the number of predictors greatly exceeds sample size.

Scientific Applications:

• Transcriptomics: Handles transcriptomic datasets with a large number of features relative to samples (P ≫ n) for predictive modeling and feature discovery.
• Biomedical Data Analysis: Applies to biomedical research problems requiring robust model evaluation and feature selection when sample size is limited relative to predictors.

Methodology:

Employs a fully nested k × l-fold cross-validation scheme with inner loops for hyperparameter tuning, outer loops for unbiased model evaluation, and filter-based feature selection nested inside the outer CV; supports glmnet-based lasso/elastic-net models and Bayesian linear/logistic regression with a horseshoe prior.