A1056
Title: Single-cell resolution deconvolution of bulk RNA-seq data via functional regression
Authors: Chong Jin - New Jersey Institute of Technology (United States) [presenting]
Xiaotian Mu - New Jersey Institute of Technology (United States)
Abstract: The proliferation of single-cell RNA sequencing (scRNA-seq) data has spurred the development of methods to deconvolve bulk RNA-seq tissues. However, many approaches rely on discrete cell type classifications, which can overlook dynamic cell states and create a false dichotomy between changes in cell composition and cell-type-specific expression. Alternatively, strategies that map bulk data to cell subpopulations may suffer from arbitrary parameter choices and "hard" classifications. BUDGIE (Bulk RNA-seq Deconvolution Using Generalized Inference and Estimation), a novel statistical framework, is proposed that deconvolves bulk RNA-seq data at the single-cell level. BUDGIE models bulk expression using a functional linear regression model. The regression coefficients represent cell abundance and are constrained to vary smoothly over a low-dimensional manifold of the scRNA-seq reference atlas. This "soft" regularization avoids rigid cell classification and captures continuous cell state transitions. Within this framework, statistical testing procedures are developed to identify significant cellular subpopulations associated with specific phenotypes. Through benchmarking against existing methods, the performance of BUDGIE is demonstrated, and it is applied to resolve cellular heterogeneity in bulk cancer tissue data.
