Analysis: Machine Learning Model Outperforms Existing Melanoma Prognostic Tools
KEY TAKEAWAYS
In a recent analysis, a machine learning-derived 14-gene panel showed robust prognostic accuracy in melanoma.
- The machine learning-derived panel outperformed 19 existing models.
Single-cell RNA sequencing suggested stronger protective gene activity in primary vs metastatic melanoma.
CUL2 was validated as a tumor-suppressive biomarker.
A new machine learning-based model may offer clinicians a more accurate method for predicting prognosis in patients with cutaneous melanoma, according to a study published in Experimental Dermatology.
Using transcriptomic data from TCGA and GEO datasets, the investigators applied 101 machine learning algorithm combinations to construct and evaluate prognostic signatures. Based on LASSO and random survival forest (RSF) methods, the optimal model demonstrated a C-index of 0.908 in the TCGA-SKCM cohort and 0.758 across four validation cohorts.
Study researchers developed and validated a 14-gene prognostic signature that outperformed 19 existing models across five independent cohorts, including data from over 600 patients with primary and metastatic melanoma.
The study authors also used single-cell RNA sequencing to assess gene expression across tumor microenvironments, finding that primary melanomas showed higher protective gene activity in CD4+ and CD8+ T cells, NK cells, and endothelial cells vs metastatic lesions. CUL2 emerged as a key protective biomarker, showing consistent downregulation in tumors and association with improved survival outcomes.
"This study screened melanoma prognosis genes from bulk melanoma transcriptomic data, built a consensus prognosis-related signature by machine learning algorithms, and verified the tumor-suppressive role of the key prognostic biomarker CUL2 by systematic molecular biological experiments," the authors concluded. "However, some limitations are as follows: The efficiency of this prognostic model has to be verified in multicentre external cohorts to ensure its generalizability. This research provides early experimental validation of the functional phenotype of CUL2 in melanoma but not its molecular mechanisms to be uncovered by subsequent investigations."
Source: Guan H, et al. Experimental Dermatology. 2025. Doi:10.1111/exd.70174