Researchers have developed a new computational tool called CADE, or Composition-Aware Differential Expression, designed to address a specific limitation in bulk transcriptomic analysis. The workflow allows scientists to assess how sensitive differential expression coefficients are to changes in cell-type composition without requiring matched single-cell reference data. This approach is particularly relevant for studies involving small cohorts or rare diseases where obtaining comprehensive single-cell references is often impractical or impossible.
Bulk differential expression analysis can produce misleading results when disease states alter the relative abundance of different cell types within a tissue sample. Existing methods that adjust for marker covariates can generate adjusted tables, but they do not directly indicate which gene-level coefficients are stable, attenuated, amplified, or representation-sensitive after these adjustments. CADE focuses on a narrower question: for each gene, how much does the reported group effect change once marker-derived composition weights are added to the model, and in which direction does that change occur?
The CADE workflow is implemented in R and estimates marker-derived relative composition weights. It fits paired unadjusted and composition-adjusted limma models to compute a per-gene Composition Confounding Index, or CCI. The tool reports signed coefficient-change labels and uses marker-dropout bootstrap rank stability to show which gene rankings persist when markers are repeatedly omitted and re-ranked. It also includes label-permutation calibration, denominator-threshold diagnostics, and isometric log-ratio coordinate-sensitivity analysis to identify genes whose classification depends on the coordinate representation of the composition weights.
In a controlled synthetic composition-bias benchmark, CADE demonstrated its ability to preserve the recovery of cell-intrinsic differential expression genes under composition bias. The tool achieved an AUROC of 0.996 at 100% synthetic bias, compared to 0.902 for unadjusted limma models.
The workflow was further evaluated using a pseudobulk analysis of macrophage activation syndrome samples. When compared against single-cell RNA sequencing-derived composition, the marker-derived weights showed a mean correlation of 0.779. All seven mappings that had a usable single-cell reference showed a p-value of less than 0.001. This evaluation covered eight preset categories, although neutrophils were excluded from this specific ground truth assessment because no reference was available for them in this context.
To illustrate the practical application of the tool, the researchers applied CADE to peripheral blood mononuclear cells from patients with familial hemophagocytic lymphohistiocytosis. The dataset included 11 patients and 33 controls. The analysis focused on a targeted panel of immune, iron, and ferroptosis-related genes.
The study emphasizes that CADE provides a reference-free and reproducible workflow for characterizing the sensitivity of bulk transcriptomic differential expression coefficients. It is important to note that the tool reports coefficient stability and does not estimate absolute cell fractions. The analysis of familial hemophagocytic lymphohistiocytosis serves as a small-cohort worked example, and the disease-specific findings are intended to be hypothesis-generating rather than definitive clinical conclusions.
By focusing on the stability of gene-level coefficients rather than deconvolution, CADE offers a distinct analytical perspective for researchers working with bulk RNA sequencing data.
Subject of Research: CADE: a reference-free workflow for coefficient-sensitivity analysis in bulk transcriptomic differential expression
Article Title: CADE: a reference-free workflow for coefficient-sensitivity analysis in bulk transcriptomic differential expression
Article References: Huang, J., Zheng, H., Jiang, H., Wei, Q., Wei, X., Wei, Y., Feng, R., & Guo, X. (2026). CADE: a reference-free workflow for coefficient-sensitivity analysis in bulk transcriptomic differential expression. BMC Bioinformatics. https://doi.org/10.1186/s12859-026-06756-x
Image Credits: AI Generated
DOI: 10.1186/s12859-026-06756-x
Keywords: CADE, reference-free, workflow, coefficient-sensitivity, analysis, bulk, transcriptomic, differential, expression, scientific research
News Source: Juliet Wilcox. (November 7, 2026). New CADE Workflow Tracks Gene Coefficient Stability in Bulk RNA Data. Scienmag.
