Clinical transcriptomics towards a companion diagnostic framework for immunotherapy in cutaneous melanoma

Melanoma remains a highly aggressive malignancy, with immunotherapy playing a pivotal role in its treatment. However, the variability in patient response highlights the need for reliable biomarkers to guide therapeutic decisions. We performed a Bioinformatics meta-analysis on a pooled dataset of 211 patients to identify clinically relevant RNA biomarkers for predicting immunotherapy response in melanoma, using bulk transcriptomics from tissue biopsies. Bulk transcriptomics remains a viable option for several reasons: projected cost per patient, feasibility in clinical settings, and turnaround time. Differential Enrichment showed a significant activation of humoral immunity, interferon-gamma-mediated responses and MHC class I binding factors in Responders. A putative blacklist of confounder genes with high fold changes was built to support future analyses. When building machine-learning-based prediction models, Random Forest Modelling based solely on transcriptomic data can predict ICB response with an AUC of 0.65, without any additional biomolecular information or clinical metadata. When adding basic clinical and technical variables the predictive power reaches an AUC of 0.68, while HLA Evolutionary Divergence brings no value in response prediction. Out-of-fold predicted response classes significantly stratify Progression-Free Survival, while Overall Survival stratification does not reach significance. Differential Deconvolution of Survival Endpoints suggests populations specific to Immune-Checkpoint blockade sensitivity and other general prognostic factors such as M1 Macrophages. Clinical Transcriptomics is herein evaluated as an additional companion biomarker tool for the current and next wave of immunotherapies in Melanoma treatment.