forgeNet

forgeNet implements a graph-embedded deep feedforward network (GEDFN) integrated with a forest-based feature graph extractor to perform supervised classification and feature selection on high-dimensional omics datasets where n ≪ p.

Key Features:

• Graph-Embedded Architecture: Incorporates known or learned functional relationships between biological units into a GEDFN to inform feature weighting and model structure and to address n ≪ p settings.
• Forest Feature Graph Extractor: Uses a forest-based ensemble classifier to learn and construct feature graphs in a supervised manner, avoiding reliance on pre-specified feature graphs.
• Supervised Learning of Feature Relationships: Integrates the forest-extracted graph with GEDFN to learn feature relationships directly from labeled omics data.
• Validated Classification Performance: Demonstrated high classification accuracy on both synthetic and real omics datasets.

Scientific Applications:

• Disease Outcome Prediction: Classification of disease outcomes from high-dimensional genomics and other omics profiles.
• Biomarker Discovery: Identification and selection of predictive features for downstream biological interpretation and validation.
• Omics Data Integration: Application to genomics, proteomics, and other omics disciplines requiring robust modeling with limited sample sizes.

Methodology:

Constructs a feature graph using a forest-based ensemble classifier and embeds that graph into a graph-embedded deep feedforward network (GEDFN) for supervised classification.