Single-Cell Transcriptomic Profiling and Machine Learning Reveal Stage-Specific Gene Dynamics in Brucellosis.

Abstract

Background: Accurate prediction of the progression of brucellosis is critical for optimizing therapeutic interventions, yet reliable biomarkers for this transition remain elusive.

Method: Single-cell RNA sequencing (scRNA-seq) of peripheral blood mononuclear cells (PBMCs) from acute (AC), subacute (SA), and chronic (CH) brucellosis patients identified nine immune subsets. Machine learning models trained on cell-type-specific gene signatures predicted brucellosis progression.

Result: Among the evaluated models, the least absolute shrinkage and selection operator showed the best classification performance, achieving an accuracy of 0.98 (95% CI, 0.98-0.99) and an AUC of 0.99 (95% CI, 0.99-1.00) for CD8+ T cells, and an accuracy of 0.96 (95% CI, 0.94-0.97) and an AUC of 0.97 (95% CI, 0.99-1.00) for NK cells. A conserved triphasic transcriptional trajectory (upregulation: healthy to AC; downregulation: AC to SA; partial recovery: SA to CH) in adaptive/innate immune cells implicated interferon signaling. Fourteen core genes (IFI44L, ISG15, and MX1) formed tightly connected networks (interaction score > 0.7), linking chronic infection to immune dysregulation.

Conclusion: We establish the first scRNA-seq-guided predictive framework for brucellosis chronicity, highlighting CD8⁺ T/NK cell signatures and interferon-driven reprogramming as therapeutic targets.