AutoDC

AutoDC applies automated machine learning to classify diseases from high-dimensional gene expression and other omics (genomic and epigenomic) data by jointly optimizing feature selection, algorithm choice, and hyper-parameter tuning.

Key Features:

• Two-Stage Feature Selection Method: Prioritizes genes with high contribution scores to filter redundant features and reduce dimensionality in gene expression datasets.
• Two-Layer Multi-Armed Bandit Framework: Uses a two-layer Multi-Armed Bandit strategy to perform simultaneous optimization across feature engineering, algorithm selection, and hyper-parameter tuning.
• Support for high-dimensional omics: Targets gene expression and other high-dimensional genomic and epigenomic data derived from next-generation sequencing.
• Benchmarking and predictive performance: Demonstrated higher predictive accuracy in disease classification on two public gene expression datasets compared with three state-of-the-art AutoML frameworks.

Scientific Applications:

• Disease classification: Classification of disease states using gene expression profiles.
• Molecular mechanism discovery: Identification of candidate genes and molecular signals underlying diseases from next-generation sequencing-based expression data.
• Genomic and epigenomic analysis: Application to high-dimensional genomic and epigenomic datasets for predictive modeling.

Methodology:

Two-stage feature selection that ranks genes by contribution scores to remove redundant features and a two-layer Multi-Armed Bandit framework that jointly optimizes feature engineering, algorithm selection, and hyper-parameter tuning.