Integrative multi-omics analysis identifies a mitochondrial dysfunction-associated diagnostic signature, molecular subtypes, and drug repurposing candidates for keloid disease.

Keloid, a fibroproliferative disorder, has limited treatments and lacks reliable biomarkers. Mitochondrial dysfunction is implicated in fibrosis, but its transcriptomic role in keloid remains incompletely characterised. We integrated five bulk transcriptomic cohorts (core training set: 46 samples; independent diagnostic validation: 7 keloid/control samples) and a single-cell RNA-seq dataset (8592 cells) to profile mitochondrial-related genes. Differential expression, WGCNA, machine learning, NMF clustering, immune infiltration, single-cell scoring, and sensitivity analyses using a stricter mitochondrial energy-metabolism subset were performed. We identified 648 mitochondrial-related differentially expressed genes, with downregulated genes enriched in cell cycle and mitotic pathways and upregulated genes enriched in immune-related pathways. A stricter mitochondrial energy-metabolism subset showed the same dominant downregulated direction (48 significant genes; 41 downregulated). WGCNA revealed ME11 as the keloid-associated module (r = 0.504, P = 3.58 ×10-4). LASSO selected a seven-gene transcriptomic signature (RAB3GAP2, NCOA6, NSF, HEBP1, IDS, MSX1, BRWD1) achieving AUC 0.881 in internal testing and 1.000 in the very small independent validation set (n = 7), which should be interpreted cautiously. The keloid-associated score (KAS), defined as an unsupervised PC1 score of these genes for stratification rather than as the supervised LASSO probability, was higher in keloid than controls (P = 1.27 × 10⁻⁵) but showed only moderate cross-cohort generalisation in GSE188952 (AUC = 0.667). Additional public GEO mining identified GSE218007 as a supportive sensitivity dataset (donor-mean AUC = 0.889), although fixed KAS performance remained heterogeneous across external datasets. Bulk immune correlations did not remain significant after FDR correction; single-cell KAS was highest in dendritic cells. Perturbation-signature analysis generated exploratory therapeutic hypotheses rather than validated drug candidates. This study provides a hypothesis-generating mitochondrial-related transcriptomic framework for keloid diagnosis and stratification. Validation in larger cohorts and functional tissue is required before the signature, KAS, or drug hypotheses are clinically actionable.