Unravelling tumour cell diversity and prognostic signatures in cutaneous melanoma through machine learning analysis.

ABSTRACT

Melanoma, a highly malignant tumour, presents significant challenges due to its cellular heterogeneity, yet research on this aspect in cutaneous melanoma remains limited. In this study, we utilized single-cell data from 92,521 cells to explore the tumour cell landscape. Through clustering analysis, we identified six distinct cell clusters and investigated their differentiation and metabolic heterogeneity using multi-omics approaches. Notably, cytotrace analysis and pseudotime trajectories revealed distinct stages of tumour cell differentiation, which have implications for patient survival. By leveraging markers from these clusters, we developed a tumour cell-specific machine learning model (TCM). This model not only predicts patient outcomes and responses to immunotherapy, but also distinguishes between genomically stable and unstable tumours and identifies inflamed ('hot') versus non-inflamed ('cold') tumours. Intriguingly, the TCM score showed a strong association with TOMM40, which we experimentally validated as an oncogene promoting tumour proliferation, invasion and migration. Overall, our findings introduce a novel biomarker score that aids in selecting melanoma patients for improved prognoses and targeted immunotherapy, thereby guiding clinical treatment decisions.