cnCV

cnCV implements consensus nested cross-validation to select stable predictive features and improve model generalizability in genomic analyses, including RNA-seq studies of major depressive disorder.

Key Features:

• Consensus Nested Cross-Validation: Combines nested cross-validation with a consensus approach that evaluates the consistency of top features across inner folds.
• Feature Stability: Prioritizes feature stability over inner-fold accuracy, reducing overfitting without requiring a predefined privacy threshold from differential privacy methods.
• Computational Efficiency: Reduces run time by eliminating the need to construct classifiers within inner folds during feature selection.
• Parsimony and Reduced False Positives: Produces more parsimonious models with fewer false positives through consensus-based feature selection.
• Implementation: Provided as an R package implementation for application to genomic and transcriptomic datasets.

Scientific Applications:

• Genomics and Bioinformatics: Applied to feature selection problems where stable predictors are critical for interpretation of complex biological datasets.
• Simulated Data Evaluation: Evaluated on simulated datasets incorporating main effects, correlations, and interactions to assess feature-selection performance.
• RNA-seq Analysis: Applied to real RNA-seq data from studies of major depressive disorder to identify robust transcriptional features.

Methodology:

Feature selection is applied within each inner fold of a nested cross-validation framework; top features across inner folds are aggregated into a consensus list to assess stability; classifiers are not constructed within inner folds, and concepts from differential privacy are integrated without requiring a privacy threshold.