TYR as a Candidate Diagnostic Biomarker for Diabetic Foot Ulcer: An Integrated Bioinformatics, Machine-Learning, and Multi-Cohort Study

Background: Diabetic foot ulcer (DFU) is a severe complication of diabetes mellitus characterized by chronic non-healing wounds and a high risk of amputation, and early molecular diagnosis remains an unmet clinical need. Methods: RNA-sequencing data from GSE134431 (13 DFU vs. eight diabetic foot-skin samples) were analyzed with a limma-equivalent empirical-Bayes moderated t-test, identifying 721 differentially expressed genes. Because the LASSO/SVM-RFE intersection was configuration-sensitive and did not recover the prioritized genes, tyrosinase (TYR) was treated as a pre-specified candidate selected by differential-expression significance, melanin-pathway enrichment, and biological plausibility; its ranks among the 721 DEGs by adjusted p-value, |log2FC|, and training AUC are reported. Performance was evaluated by nested leave-one-out cross-validation (LOOCV), apparent AUC with bootstrap 95% confidence intervals (CIs), and external cohorts (GSE80178 and GSE68183; quantile-normalized and z-scored per dataset). Single-cell data (GSE165816) were re-analyzed to quantify TYR before and after quality control, and CIBERSORT was used for immune deconvolution with Benjamini–Hochberg correction and group-adjusted correlations. Results: In nested LOOCV, TYR was re-selected in all 21 folds with a held-out AUC of 0.971. TYR was significantly downregulated in the external cohorts (merged six DFU vs. 12 non-DFU: p = 0.0047, direction-aware AUC = 0.903, bootstrap CI 0.736–1.000), including the within-GSE80178 comparison of DFU versus diabetic foot skin (p = 0.0238). In single-cell analysis, TYR was genuinely expressed after quality control (64.2% of melanocytes), and melanocyte proportions were reduced in DFU at the sample level (p = 0.040). After correction for multiple testing, resting and activated mast cells, CD4+ memory-activated T cells, and follicular helper T cells differed between DFU and controls, and TYR showed a group-adjusted positive correlation with M1 macrophages (R = 0.460, p = 0.036). Conclusions: TYR is a reproducible candidate diagnostic biomarker for DFU supported by nested validation, external cohorts, and single-cell evidence. HOXC10, which was not replicated externally, is retained as an exploratory observation only. Larger prospective cohorts with diabetic non-ulcer controls and functional experiments are required.