New algorithm boosts gene expression marker detection across diverse biological systems

Researchers at the Germans Trias i Pujol Research Institute (IGTP), Universitat Politècnica de Catalunya (UPC), IrsiCaixa, and CIBEREHD have introduced a computational strategy to select gene signatures that characterize cellular states more precisely from mRNA-seq data. Their work appears in Frontiers in Immunology and targets a central challenge in transcriptomics: identifying informative genes that faithfully reflect dynamic biology.

During infection, inflammation, or tumor progression, cells alter the expression of thousands of genes. These changes form high-dimensional molecular landscapes that researchers use to infer how cells transition between functional states. Yet conventional gene-selection pipelines—often reliant on differential expression assumptions—may miss key structure in the data, especially when states evolve over time or differ across experimental contexts.

To overcome these limitations, the team developed the Cartesian Distance-Based Gene Expression (CDBGE) approach. Instead of focusing solely on expression differences between conditions, CDBGE uses distance-based comparisons to determine which genes best separate biological scenarios. This allows the algorithm to incorporate multidimensional relationships among samples and align them with biologically meaningful groupings.

The method was validated across multiple publicly available datasets spanning human and mouse systems. Importantly, evaluations covered heterogeneous experimental designs, testing whether CDBGE can generalize beyond a single dataset or experimental protocol. Across these benchmarks, the approach both classified samples effectively and produced gene markers that remained informative under diverse conditions.

A key feature of CDBGE is its ability to integrate temporal and multidimensional information into gene selection. This improves interpretability relative to more opaque feature-ranking strategies, while also expanding the range of potential biomarkers beyond those that are already well established.

As first author Qiaoling Ye notes, unlike standard differential expression analyses, CDBGE better captures the complexity of biological behavior by leveraging richer structure in the data. This enables discovery of both known and previously unrecognized biomarkers tied to cellular heterogeneity.

The researchers also emphasize that CDBGE remains simple and flexible, enabling its use across varied study designs in computational biology and immunology. By improving the accuracy of gene selection without sacrificing interpretability, the framework could support more reliable biomarker discovery.

More broadly, the work provides a distance-based lens for mapping state transitions in complex cellular systems. As transcriptomic studies increasingly probe time-resolved and heterogeneous processes, approaches like CDBGE may help translate high-throughput data into clearer biological narratives.