DART

DART estimates pathway activation in tumor gene-expression datasets by evaluating consistency between prior molecular signatures (e.g., in-vitro perturbation expression data) and inferred network topology to infer activation status of molecular pathways across cancer samples.

Key Features:

• Network-based methodology: Employs a network-based approach that leverages gene expression signatures from perturbation experiments and structural models of molecular interactions.
• Consistency evaluation: Evaluates consistency between inferred pathway networks from cancer expression data and prior model signatures.
• Modular structure analysis: Identifies modular structure within pathway networks to capture interacting pathway modules.
• Pruning network strategy: Applies a pruning strategy to network topology to infer activation status of molecular signatures in individual samples.
• Boolean interaction Cox-regression models: Uses Boolean interaction Cox-regression models for prognostic subtype classification.

Scientific Applications:

• Breast cancer cohort analysis: Applied to a panel of 438 estrogen receptor-negative (ER-) and 785 estrogen receptor-positive (ER+) breast cancer samples to identify pathway activation patterns.
• Immune response modulation: Identified antagonistic roles between T-cell helper-1 (Th1) and T-cell helper-2 (Th2) immune-response modules as major risk factors for distant metastasis in ER- basal and HER2+ breast cancers.
• Prognostic subtype classification: Demonstrated that high Th1 activation combined with low TGF-beta pathway activity correlates with particularly favorable prognosis and outperforms classifications based on individual pathways.
• Synergistic pathway effects: Revealed that simultaneous high MYC and RAS activity in ER+ breast cancer is associated with significantly worse prognosis than either factor alone.
• Independent validation: Prognostic classifications were validated in independent datasets comprising 173 ER- and 567 ER+ breast cancers.

Methodology:

DART applies a network-based approach that leverages gene expression signatures from perturbation experiments and structural molecular-interaction models, evaluates consistency between inferred pathway networks and prior model signatures, detects modular network structure, applies a network-pruning strategy to infer signature activation in individual samples, and employs Boolean interaction Cox-regression models for prognostic subtype classification.