DCNet

DCNet infers cellular composition from bulk RNA-Seq data using an explainable deep neural network to characterize tumor microenvironment (TME) heterogeneity.

Key Features:

• Deep Learning-Based Framework: DCNet employs an explainable artificial neural network architecture incorporating visible design principles for model interpretability.
• Integration of Cell-Marker Relationships: The model embeds relationships between cell types and their marker genes within the neural network framework.
• High-Resolution Cellular Inference: DCNet infers a cellular landscape comprising over 400 distinct cell types from bulk RNA-Seq data.
• Robustness and Stability: The model accurately recapitulates cell landscapes in multiple single-cell RNA-Seq datasets, demonstrating robustness and stability.
• Clinical Relevance: Applied to TCGA patient data, DCNet stratifies patients into groups with significant differences in survival times and distinct cellular compositions.

Scientific Applications:

• Tumor Microenvironment Characterization: Provides detailed cellular composition of TMEs to study how cellular heterogeneity influences tumor behavior.
• Immunotherapy Research: Identifies patient subgroups with distinct immune profiles that may predict responses to immunotherapy.
• Clinical Stratification and Survival Analysis: Enables stratification of TCGA cohorts by cellular composition to investigate associations with patient survival.

Methodology:

DCNet trains an explainable deep neural network on bulk RNA-Seq data while explicitly modeling cell type–marker gene relationships and validates inferred compositions against single-cell RNA-Seq datasets.