EnsDeconv

EnsDeconv performs ensemble deconvolution to estimate cellular fractions from tissue-level omics data, enabling extraction of cell-type-specific signals from transcriptomics and epigenomics datasets.

Key Features:

• Ensemble deconvolution: Integrates outputs from multiple single-cell deconvolution methods and synthesizes them using cell-type-specific (CTS) robust regression.
• Model diversity: Combines results from 11 distinct deconvolution techniques, 10 reference datasets, 5 marker gene selection procedures, 5 data normalization strategies, and 2 transformations.
• Evaluation on real data: Rigorously evaluated on four large real datasets comprising 4,937 tissue samples with measured cellular fractions and bulk gene expression across various tissues.
• Multi-omics versatility: Applicable to transcriptomics (RNA microarray and RNA-seq) and extended to bulk DNA methylation data.
• Support for downstream CTS analyses: Produces cellular fraction estimates that enable analyses such as differential cell-type fraction analysis in relation to clinical variables.
• Empirical data sources: Includes application to RNA microarray data from the TRAUMA study (GSE36809) and to RNA-seq datasets.

Scientific Applications:

• Differential analysis: Investigating how cellular composition varies with clinical parameters through differential cell-type fraction analysis.
• Multi-omics integration: Integrating transcriptomic and epigenomic (bulk DNA methylation) signals to study cell-type-specific effects.
• Translational research: Linking cell-type compositions from bulk samples to disease states or treatment responses for clinical and biomarker studies.

Methodology:

Ensemble deconvolution that integrates multiple single-cell deconvolution methods and synthesizes outputs via CTS robust regression, leveraging 11 deconvolution techniques, 10 reference datasets, 5 marker selection procedures, 5 normalization strategies, and 2 transformations; evaluated on four real datasets totaling 4,937 samples with measured cellular fractions and bulk gene expression and extended to bulk DNA methylation data.