DecompPipeline

DecompPipeline performs reference-free deconvolution preprocessing and feature selection of bulk DNA methylation data to recover cell type–specific methylation signals and latent methylation components (LMCs) from heterogeneous tissues.

Key Features:

• Preprocessing and Feature Selection: Adjusts for confounders using independent component analysis (ICA) and selects methylation features for downstream deconvolution.
• Integration with Deconvolution Methods: Interfaces with MeDeCom, RefFreeCellMix, and EDec to enable retrieval of latent methylation components (LMCs) from bulk methylomes.
• Guided Biological Inference: Supports interpretation and validation of deconvolution results using FactorViz to aid identification of cell-type components.

Scientific Applications:

• Population-scale methylation analysis: Enables cell type–specific methylation studies across large patient cohorts when single-cell data are unavailable.
• Tumor heterogeneity and cellular composition: Facilitates dissection of cellular heterogeneity in complex tissues, including identification of tumor and stromal methylation signatures.
• TCGA lung cancer analyses: Applied to TCGA lung cancer methylomes to identify proportions of stromal cells and tumor-infiltrating immune cells and their associations with clinical parameters.

Methodology:

The protocol comprises three computational stages: data preprocessing (confounder adjustment via ICA and feature selection by DecompPipeline), deconvolution (application of MeDeCom, RefFreeCellMix, and EDec to retrieve LMCs), and biological inference and validation (interpretation using FactorViz).