hydra_bnpy

hydra_bnpy applies Bayesian non-parametric statistics to perform unsupervised clustering of gene expression and identify multimodal expression signatures for pediatric cancer subtyping without requiring matched normal tissue.

Key Features:

• Unsupervised clustering: Applies unsupervised clustering to gene expression measurements to identify multimodally distributed genes and differential pathway expressions.
• Dirichlet process mixture model: Implements a Bayesian non-parametric Dirichlet process mixture model to automatically detect complex multimodal patterns in gene expression data.
• Matched-normal-independent analysis: Identifies expression signatures and subtypes without requiring matched normal tissue samples.
• Increased sensitivity: Demonstrates higher sensitivity than traditional gene set enrichment methods for detecting multimodal expression signatures.
• Tumor-type applicability: Applied to small blue round cell tumors, including rhabdomyosarcoma, synovial sarcoma, neuroblastoma, Ewing sarcoma, and osteosarcoma, to detect immune and stromal expression changes.
• Association detection: Identifies genotype–phenotype associations such as ATRX deletions correlated with elevated immune marker expression in high-risk neuroblastoma.
• Cross-tumor subtype discovery: Reveals similar subtypes across diverse small blue round cell tumors to support comparative oncology analyses.

Scientific Applications:

• Pediatric cancer subtyping: Dissects molecular subtypes in pediatric cancers using multimodal gene expression signatures.
• Tumor microenvironment profiling: Characterizes immune and stromal alterations associated with specific expression-based subtypes.
• Biomarker and association discovery: Uncovers biomarkers and genotype–expression associations relevant to prognosis and therapeutic response, exemplified by ATRX-related findings.
• Comparative oncology and extension to adult malignancies: Enables cross-tumor comparisons and can be applied to adult cancers with similar gene expression analysis challenges.

Methodology:

Uses Bayesian non-parametric statistics via a Dirichlet process mixture model for unsupervised clustering of gene expression to detect multimodally distributed genes and differential pathway expressions without requiring matched normal tissue.