Single-cell and machine learning integration reveals ferroptosis-driven immune landscapes for melanoma stratification

TL;DR

This study uses single-cell data and machine learning to identify ferroptosis-driven immune patterns in melanoma, enabling better patient stratification and treatment guidance.

Contribution

A novel multi-omics framework integrating ferroptosis and immune signatures with machine learning for melanoma stratification and prognosis.

Findings

Three ferroptosis-immune subtypes with distinct survival and immune profiles were identified.

A 40-gene prognostic signature effectively stratified patient survival risk and predicted chemotherapy sensitivity.

Single-cell analysis linked ferroptosis activity to immunosuppressive microenvironments via POSTN–ITGB5 signaling.

Abstract

Ferroptosis, a regulated form of cell death, has emerged as a critical modulator of melanoma's tumor progression and immune evasion. However, its integration with the tumor immune microenvironment (TME) and clinical prognostication remains underexplored. This study aims to construct a multi-omics framework combining ferroptosis-related signatures, immune infiltration patterns, and machine-learning approaches to stratify melanoma patients and guide therapeutic decision-making.

We developed a multi-omics framework integrating bulk transcriptomics (TCGA/GEO), single-cell RNA sequencing, and machine learning to decode melanoma's ferroptosis-immune axis. Ferroptosis-immune subtypes were identified through consensus clustering and immune profiling, while prognostic models were constructed via LASSO/stepwise Cox regression and machine learning optimization.

Three ferroptosis-immune subtypes exhibiting distinct survival outcomes and immune phenotypes were identified. A 40-gene prognostic signature (externally validated) effectively stratified patient survival risk and predicted chemotherapy sensitivity. Single-cell analysis revealed elevated ferroptosis activity within an immunosuppressive microenvironment, specifically implicating POSTN–ITGB5 signaling in fibroblast-immune cell crosstalk. A clinically applicable nomogram integrating risk scores and clinical factors demonstrated robust predictive accuracy (AUC 0.829–0.845). Machine learning refined a 4-gene prognostic signature (CLN6, GMPR, AP1S2, ITGA6), with functional validation confirming the role of CLN6 in proliferation and migration.

This study establishes a prognostic framework and therapeutic roadmap for precision immuno-oncology in melanoma, bridging multi-omics discovery with clinical translation.

Figures