Answers are grounded in this paper and cite their sources.
Background Diabetic nephropathy (DN) is a leading cause of end-stage renal disease, and available markers incompletely capture disease activity. Protein glycosylation is increasingly implicated in diabetic kidney injury, but the cross-platform reproducibility of glycosylation-centric transcriptional modules remains unclear. Methods A protein-coding glycosylation-related gene (GRG) list (n = 812; Gene Ontology and UniProt glycosylation terms, filtered against NCBI Gene type_of_gene annotation) was constructed. Differentially expressed genes (DEGs) in glomeruli from the discovery cohort GSE96804 (41 DN vs. 20 controls) were intersected with GRGs. Two exploratory LASSO panels (performance-first lambda.min; parsimony-first lambda.1se) were derived by univariate logistic regression, random forest, and LASSO, and assessed by ROC analysis, calibration, decision curve analysis, and leak-free repeated cross-validation with every supervised step re-estimated within each fold. External validation used GSE30528 (glomeruli, 9 DKD vs. 13 controls) and GSE30529 (tubulointerstitium, 10 DKD vs. 12 controls). Immune infiltration was profiled by nu-support-vector-regression (nu-SVR) deconvolution of the official LM22 matrix. Results Of 610 DEGs, 27 protein-coding GRDEGs (15 up-, 12 down-regulated) remained after the coding-gene filter. GRDEGs were markedly enriched in glycosaminoglycan and heparin binding, extracellular matrix, and integrin/focal adhesion signalling. The 7-gene lambda.min panel outperformed the 5-gene lambda.1se panel on all four evaluation layers. The apparent training AUC was 0.9988 (saturated); leak-free out-of-fold AUCs were 0.9565 ± 0.0210 (lambda.min; optimism 0.0423) and 0.9561 ± 0.0236 (lambda.1se; optimism 0.0415). External AUCs were 0.778 (p = 0.033) and 0.683 (p = 0.156) for the lambda.min panel and 0.735 (p = 0.071) and 0.625 (p = 0.339) for the lambda.1se panel. Neither composite score exceeded its best single-gene member in either cohort (best single genes: LPL 0.983, FN1 0.923, EGF 0.967, FN1 0.958). FN1 was up-regulated and significant in all three cohorts (AUC 0.911/0.923/0.958). SELL tracked the neutrophil signal at the expression layer (r ≈ 0.84) and the proportion layer (r = 0.31–0.47 across all three cohorts). Conclusions We characterize a glycosylation-centric ECM–GAG transcriptional module in DN with FN1 as a cross-cohort hub. The composite risk scores were exploratory, did not outperform single genes, and are not a validated diagnostic model; prospective multi-cohort validation is required.
No takes yet. Share an insight, caveat, or question.
Chen et al. (2026) studied this question.