Integrated single-cell and bulk RNA-Seq analysis identifies ACSL4-related ferroptosis-senescence signatures to predict prognosis in melanoma via machine learning

Cutaneous melanoma is characterized by high cellular heterogeneity and unpredictable clinical outcomes. Emerging evidence suggests a complex crosstalk between ferroptosis and cellular senescence in reshaping the tumor microenvironment (TME). However, an integrated prognostic framework leveraging both single-cell and bulk transcriptomics via advanced machine learning remains to be established. Single-cell RNA sequencing (scRNA-seq) profiles of melanoma were analyzed to map TME heterogeneity and calculate cellular ferroptosis-senescence (FS) activity via the AUCell algorithm. Core hub genes bridging scRNA-seq and macro-cohorts were identified using Weighted Gene Co-expression Network Analysis (WGCNA). To build a robust model without data leakage, dozens of machine learning pipelines were trained and cross-validated strictly within The Cancer Genome Atlas Cutaneous Melanoma cohort (TCGA-SKCM) and independently validated across three external GEO datasets (GSE19234, GSE22153, GSE65904). Loss-of-function assays (colony formation and Transwell migration) were deployed to validate the biological role of the pivotal gene, ACSL4, in A375 cells. The scRNA-seq landscape distinguished 20 cell clusters, within which ACSL4 emerged as the top gene positively correlating with the FS score. WGCNA identified the MEturquoise module as highly associated with the FS trait, yielding 42 candidate hub genes. An ensemble machine learning model based on Elastic Net achieved the highest mean C-index across cohorts, outperforming previously published signatures. High-risk patients defined by this signature exhibited significantly truncated overall survival. In vitro experiments confirmed that silencing ACSL4 significantly suppressed melanoma cell proliferation and migration. We established and validated a novel, single-cell-informed ferroptosis-senescence prognostic classifier using an optimized machine learning framework, while identifying ACSL4 as a key oncogenic driver in melanoma.