Machine learning–guided senescence signatures identify shikonin as an anti-inflammatory therapy for keloid

Background Keloid is a chronic inflammatory disorder resulting from dysregulated wound healing, characterized by tumor-like aggressiveness and a high recurrence rate. Current treatment modalities remain unsatisfactory. Cellular senescence (CS) and the senescence-associated secretory phenotype (SASP) are persistently present in keloid and may promote disease progression by secreting pro-inflammatory factors and inducing senescence in neighboring cells; however, the underlying regulatory mechanisms remain largely unexplored. Shikonin, a major bioactive component from traditional Chinese medicine, exhibits anti-proliferative, anti-inflammatory, and pro-apoptotic activities, but its therapeutic potential in keloid and its relationship with cellular senescence require further investigation. Technology We profiled CS in keloid using GEO transcriptomes supplemented by RNA sequencing. Differentially expressed genes (DEGs) were identified, and CS-associated DEGs (CS-DEGs) were annotated via the CellAge database, followed by GO/KEGG enrichment analyses. To prioritize clinically relevant CS-DEGs, we applied an AI-driven machine learning (ML) gene-prioritization pipeline to identify hub biomarkers/targets. Immune microenvironment composition was quantified using CIBERSORT algorithmic deconvolution, estimating 22 immune cell subsets from bulk expression data. Subsequently, in vitro and ex vivo organ culture experiments were conducted to validate the effects of shikonin on the senescent phenotype, SASP factor expression, and apoptosis in keloid fibroblasts (KFs), supported by in silico molecular docking to evaluate shikonin–target interactions. Results We identified 187 CS-DEGs enriched in inflammation, stress responses, cell-cycle regulation, tumorigenesis, and tumor suppression. ML highlighted AGR2 (anterior gradient 2) and MDK (midkine) as hub genes with strong diagnostic performance; AGR2 was downregulated and MDK upregulated in keloid and validated in an independent dataset. CIBERSORT revealed increased macrophages M1, which correlated with hub CS-DEGs, supporting a senescence–inflammation axis in keloid. Experimentally, keloid tissues showed increased senescence-associated β-galactosidase–positive cells and elevated MDK, TGF-β1, IL-1β, IL-6, IL-8, and P16 expression versus normal skin. Shikonin inhibited KFs proliferation in a time- and dose-dependent manner; reduced MDK, collagen I, TGF-β1, VEGF, IL-1β, IL-6, IL-8, PPARγ, NOTCH1, and BCL-2; and increased AGR2 and BAX in cultured KFs. In silico molecular docking predicted favorable binding affinity between shikonin and target proteins. In vitro , shikonin suppressed KFs outgrowth from tissue masses and promoted apoptosis. Collectively, these findings support a CS-associated, inflammation-coupled mechanism in keloid, nominate AGR2 and MDK as candidate biomarkers/targets, and position shikonin as a candidate therapy targeting inflammation, senescence and apoptosis resistance, illustrating the value of AI/ML-enabled biomarker discovery integrated with algorithmic immune deconvolution and experimental validation.