Document Type : Original Article
Authors
1 Taleghani General Hospital, Division of Nephrology, Department of Internal Medicine, Shahid Beheshti University of Medical Sciences, Tehran, Iran.
2 Clinical Research and Development Center, Division of Nephrology, Department of Internal Medicine, Shahid Modarres Hospital, Shahid Beheshti University of Medical Sciences, Tehran, Iran.
3 Skull Base Research Center, Loghman Hakim Hospital, Shahid Beheshti University of Medical Sciences, Tehran, Iran.
Abstract
Background: Renal fibrosis represents the final common pathway of chronic kidney disease (CKD); however, both its definitive diagnostic biomarkers and the principal cellular mediators driving its progression remain incompletely characterized.
Objective: To derive a machine-learning-based transcriptomic signature from single-cell RNA-sequencing data that predicts kidney fibrosis severity and elucidates the underlying immune-stromal cellular interactions.
Methods: A machine-learning approach incorporating Random Forest and Least Absolute Shrinkage and Selection Operator (LASSO) regression was used to identify a sparse transcriptomic signature. Cell-cell communication networks and pseudotime trajectories were reconstructed to characterize the fibrotic niche. The five-gene signature (CXCL13, CCL19, TNFSF13B, IL6, and TGFB1) was then correlated with eGFR, UACR, and Banff scores, and externally validated in the independent human kidney scRNA-seq dataset GSE183276.
Results: The fibrotic kidney microenvironment exhibited a marked expansion of fibroblasts and CD19+ B-cells. Machine-learning feature selection identified a highly predictive five-gene molecular signature comprising CXCL13, CCL19, TNFSF13B (encoding BAFF), IL6, and TGFB1. Intercellular network analysis revealed dominant B-cell–fibroblast signaling that was significantly associated with fibroblast transdifferentiation into extracellular matrix-producing myofibroblasts. The signature score correlated with eGFR, UACR, and Banff scores. External validation in the independent dataset GSE183276 confirmed the diagnostic robustness of this signature.
Conclusions: We present a validated machine-learning -derived transcriptomic signature that reflects the immune-stromal dynamics of renal fibrosis. This five-gene signature accurately predicts CKD severity and, importantly, implicates B-cell-mediated fibroblast activation as a promising diagnostic biomarker and a potential therapeutic target.
Keywords
)