DeCompress: tissue compartment deconvolution of targeted mRNA expression panels using compressed sensing

Abstract

Targeted mRNA expression panels, measuring up to 800 genes, are used in academic and clinical settings due to low cost and high sensitivity for archived samples. Most samples assayed on targeted panels originate from bulk tissue comprised of many cell types, and cell-type heterogeneity confounds biological signals. Reference-free methods are used when cell-type-specific expression references are unavailable, but limited feature spaces render implementation challenging in targeted panels. Here, we present DeCompress, a semi-reference-free deconvolution method for targeted panels. DeCompress leverages a reference RNA-seq or microarray dataset from similar tissue to expand the feature space of targeted panels using compressed sensing. Ensemble reference-free deconvolution is performed on this artificially expanded dataset to estimate cell-type proportions and gene signatures. In simulated mixtures, four public cell line mixtures, and a targeted panel (1199 samples; 406 genes) from the Carolina Breast Cancer Study, DeCompress recapitulates cell-type proportions with less error than reference-free methods and finds biologically relevant compartments. We integrate compartment estimates into cis-eQTL mapping in breast cancer, identifying a tumor-specific cis-eQTL for CCR3 (C–C Motif Chemokine Receptor 3) at a risk locus. DeCompress improves upon reference-free methods without requiring expression profiles from pure cell populations, with applications in genomic analyses and clinical settings.

Multiomics in primary and metastatic breast tumors from the AURORA US network finds microenvironment and epigenetic drivers of metastasis

“…5a ), including signatures of fibroblasts/stromal cells and endothelial cells and many adaptive immunity signatures as being lower in metastases. However, when supervised analyses were performed within a gene expression subtype, which is known to associate with the likelihood of metastasis 33 , 34 , then subtype-specific differences were observed (Fig. 2c,d ).…”

“…5a), including signatures of fibroblasts/stromal cells and endothelial cells and many adaptive immunity signatures as being lower in metastases. However, when supervised analyses were performed within a gene expression subtype, which is known to associate with the likelihood of metastasis 33,34 Significance of the differences between primary tumors and metastases were calculated using LMMs (q < 0.01). Significant signatures are row ordered from high to low according to β-coefficients (or regression coefficients) and divided according to upregulated (positive) or downregulated (negative) in metastasis.…”

Molecular Group and Correlation Guided Structural Learning for Multi-Phenotype Prediction

“…The predicted cell type fractions from Brilliant were compared with other commonly used reference-based cell-type deconvolution methods, as in the study by Cai et al [7], including CIBERSORT/CIBERSORTx [25], DeconRNASeq [13], DCQ [2], DeCompress [5], DSA [36], dtangle [17], EnsDeconv (Ensemble Deconvolution) [7], GEDIT (Gene Expression Deconvolution Interactive Tool) [23], EPIC (Estimating the Proportions of Immune and Cancer cells) [27], FARDEEP (Fast And Robust DEconvolution of Expression Profiles) [14], hspe (hybrid-scale proportions estimation) [18], ICeDT (Immune Cell Deconvolution in Tumor tissues) [33], and SCDC methods including non-negative least squares (SCDC NNLS) and inverse sum of squares (SCDC ISSE) [9]. All the results were based on the LM22 reference.…”

Multiplatform Analysis of Primary and Metastatic Breast Tumors from the AURORA US Network identifies microenvironment and epigenetics drivers of metastasis

“…4a) including signatures of fibroblasts/stromal cells, endothelial cells, and many adaptive immunity signatures as being lower in metastasis. However, when supervised analyses are performed within a gene expression subtype, which is known to associate with the likelihood of metastasis to a given site 27,28 , then subtype-specific differences are observed (Fig. 2c, d).…”