DeCompress

DeCompress deconvolves targeted mRNA expression panels into tissue compartments to estimate cell-type proportions and identify compartment-specific gene signatures for downstream genomic analyses.

Key Features:

• Semi-Reference-Free Approach: Operates in a semi-reference-free manner, enabling deconvolution when pure cell-type reference profiles are unavailable.
• Reference Data Leveraging: Leverages reference RNA-seq or microarray datasets from similar tissues to augment information for targeted panels.
• Feature Space Expansion via Compressed Sensing: Applies compressed sensing to expand the feature space of targeted mRNA panels (typically up to 800 genes) to improve deconvolution resolution.
• Ensemble Reference-Free Deconvolution: Performs ensemble reference-free deconvolution on the expanded dataset to estimate cell-type proportions and identify gene signatures.
• Validation and Performance: Demonstrated superior accuracy over traditional reference-free methods in simulated mixtures, public cell line mixtures, and a Carolina Breast Cancer Study targeted panel (1199 samples, 406 genes) while recovering biologically relevant compartments.
• Integration with Genomic Analyses: Enables incorporation of compartment estimates into genomic analyses, exemplified by cis-eQTL mapping and discovery of a tumor-specific cis-eQTL for CCR3 at a risk locus.

Scientific Applications:

• Deconvolution of Targeted Panels: Deconvolving targeted mRNA expression panels, including archived samples and assays with limited gene sets.
• Cell-Type Proportion Estimation: Estimating relative abundances of tissue compartments in bulk tissue samples to correct for cellular heterogeneity.
• Compartment-Specific Signature Discovery: Identifying compartment-specific gene signatures for interpretation of complex tissue biology.
• Genomic Association Studies: Incorporating compartment estimates into analyses such as cis-eQTL mapping in breast cancer to reveal compartment-specific genetic effects (e.g., CCR3 tumor-specific cis-eQTL).

Methodology:

Leverages reference RNA-seq or microarray datasets to expand the feature space of targeted panels via compressed sensing, then applies ensemble reference-free deconvolution on the expanded dataset to estimate cell-type proportions and identify gene signatures.