Background:
Methylated circulating cell-free DNA (m-cfDNA) offers a promising avenue for non-invasive cancer biomarker development. This study aimed to identify a panel of functional melanoma-specific methylation sites that can be used as biomarkers for monitoring treatment response in cutaneous melanoma through m-cfDNA analysis.
Method:
Cutaneous melanoma-specific differentially methylated sites were identified from The Cancer Genome Atlas (TCGA) using a one-vs-all approach based on Δβ and adjusted p-values across five tumour types (Cutaneous melanoma, Lung, Liver, Colon, Uveal melanoma). Sites were subsequently filtered to remove m-cfDNA background signal using blood cells methylation dataset (GSE87571). Next, a three-step feature selection approach (ReliefF, K-means clustering, and methylation-expression integrative analysis) was used to refine and keep functionally relevant sites. Finally, a supervised machine learning method, PLS-DA, was employed to develop a prediction model and selected markers were validated externally (GSE51547 and GSE44661).
Results:
We identified 50 cutaneous melanoma-specific differentially methylated sites achieving AUC, sensitivity, and specificity >85% in distinguishing cutaneous melanoma from other cancer types. Sixteen candidates showed correlations with gene expression in external datasets. The 3 highest-performing markers SHOX2 (cg09095222), ALDH2 (cg07106761), KIFC2 (cg05317714) were prioritised for validation in patient samples.
Conclusion:
Our findings provide a panel of biomarker candidates suitable for monitoring treatment response in cutaneous melanoma patients.