A versatile distance-based approach for gene expression selection across diverse biological systems.

Differential gene expression analysis is essential for characterizing immune cell phenotypes, yet conventional approaches-typically based on log2 fold-change (log2FC) and False Discovery Rate (FDR) thresholds-often struggle to capture the complexity and continuum of transcriptional states. To address this limitation, we developed a new computational method for gene selection from mRNA-seq data: the Cartesian Distance-Based Gene Expression (CDBGE) selector. This algorithm identifies differentially expressed genes by leveraging multidimensional expression distances rather than relying on traditional univariate statistical cutoffs, enabling a more refined and biologically coherent gene-marker selection. We applied the CDBGE selector to construct a gene-based framework for distinguishing macrophage polarization states. The model was trained using publicly available macrophage transcriptomic datasets and subsequently validated with in vitro human macrophages stimulated with IFN-γ/LPS, conditioned medium from HepG2 liver cancer cells (Sec-HepG2), or IL10. To evaluate its generalizability beyond macrophage biology, we further tested the method on human embryonic stem cell differentiation datasets. Compared with standard differential expression pipelines, the CDBGE selector more effectively identified subtype-specific markers and revealed dynamic transcriptional transitions over time. These findings demonstrate that distance-based gene selection provides an improved strategy for analyzing complex mRNA-seq datasets. Overall, the CDBGE selector offers a robust, scalable, and broadly applicable tool for differential gene expression analysis and phenotype characterization.