IECP: iterative equilibration of cell-type expression profiles improves accuracy of reference-free deconvolution.

Abstract

Reference-free deconvolution methods are widely used to estimate cell-type composition and expression profiles from bulk transcriptomic data when native cell type references are unavailable. However, these methods often suffer from systematic biases caused by the asymmetric differential gene expressions across cell types and biological or experimental conditions, producing inaccurate proportion inference and reduced interpretability. Here we present IECP (Iterative Equilibration of Cell-type Expression Profiles), an R package that improves deconvolution accuracy by iteratively equilibrating the asymmetric differential gene expressions across cell types. IECP identifies consistently expressed genes (CEGs) across estimated cell-type profiles, computes CEG-based sample-wise scaling factors, and equilibrates the bulk data matrix before the next deconvolution iteration. By integrating IECP with five popular reference-free deconvolution methods, CAM3.0, TOAST, PREDE, RefFreeEWAS, and CDseq, we demonstrate consistent improvements in cell-type proportion estimation on multiple benchmark datasets. IECP R package is freely available at https://github.com/niccolodpdu/IECP, with sample data and application vignettes. Supplementary data are available at Bioinformatics online.