EpiDISH

EpiDISH deconvolutes cell-type proportions in mixed tissue samples using DNA methylation data and DHS-informed reference profiles to correct for intra-sample cellular heterogeneity in Epigenome-Wide Association Studies (EWAS).

Key Features:

• DHS-informed reference database: Incorporates cell-type specific DNAse Hypersensitive Site (DHS) information from the NIH Epigenomics Roadmap to construct an improved reference DNA methylation database.
• Inference methods: Implements Robust Partial Correlations (RPC), CIBERSORT (CBS), and Constrained Projection (CP) for cell-type proportion estimation.
• Reference-based algorithms: Utilizes predefined DNA methylation profiles of relevant cell types for deconvolution rather than reference-free approaches.
• Improved accuracy for blood and epithelial mixtures: Produces statistically improved cell-count estimates for whole blood and epithelial cell mixtures relative to traditional references.
• Robustness to noise and tissue variability: RPC combined with DHS information has been shown to yield robust performance across multiple tissue types and realistic noise levels.
• Validation with purified samples: Supports validation of inferred cell distributions using external validation sets from purified leukocyte samples.

Scientific Applications:

• EWAS covariate correction: Adjusts for cell-type composition in Epigenome-Wide Association Studies to reduce confounding from cellular heterogeneity.
• Leukocyte composition quantification: Quantifies leukocyte subset distributions in whole blood to detect differences between cases and controls.
• Tumor microenvironment characterization: Deconvolutes epithelial and immune cell mixtures in cancer datasets, including Head and Neck Squamous Cell Carcinoma (HNSCC) and ovarian cancer.
• Large-scale immunological studies: Provides cell-type proportion estimates for studies investigating immune composition and disease mechanisms.

Methodology:

EpiDISH uses reference-based DNA methylation deconvolution combining DHS-derived reference profiles from the NIH Epigenomics Roadmap with inference methods Robust Partial Correlations (RPC), CIBERSORT (CBS), and Constrained Projection (CP), and validates estimates against purified leukocyte sample sets.