Oral Presentation Australasian Melanoma Conference 2026

Predicting who benefits from immunotherapy in advanced melanoma: a blood-based multi-omic signature (#76)

Nurudeen A. Adegoke 1 2 3 , Russell J. Diefenbach 1 4 , Ashleigh Stewart 1 4 , Aaron B. Beasley 5 6 , Wei Yen Chan 1 4 , Ines Pires da Silva 1 2 3 , Anna Reid 5 6 , Lydia Warburton 2 5 , Alexander M. Menzies 1 2 7 , Michael J. Millward 8 9 , Serigne N. Lo 1 2 3 , Elin S. Gray 5 6 , Georgina V. Long 1 2 3 7 , Helen Rizos 1 4
  1. Melanoma Institute Australia, The University of Sydney, Sydney, NSW, Australia
  2. Faculty of Medicine and Health, The University of Sydney, Sydney, NSW, Australia
  3. Charles Perkins Centre, The University of Sydney, Sydney, NSW, Australia
  4. Faculty of Medicine, Health and Human Sciences, Macquarie University, Sydney, NSW, Australia
  5. Centre for Precision Health, Edith Cowan University, Joondalup, WA, Australia
  6. School of Medical and Health Sciences, Edith Cowan University, Joondalup, WA, Australia
  7. Department of Medical Oncology, Northern Sydney Cancer Centre, Royal North Shore Hospital, Sydney, NSW, Australia
  8. Faculty of Health and Medical Sciences, The University of Western Australia, Perth, WA, Australia
  9. Department of Medical Oncology, Sir Charles Gairdner Hospital, Perth, WA, Australia

Background. About half of patients with advanced melanoma derive durable benefit from immune checkpoint inhibition (ICI), while others gain little and remain at risk of toxicity and cost. A minimally invasive blood test identifying likely benefit would improve patient selection. No single circulating biomarker is sufficiently accurate, however, and response appears to depend on several biological layers, providing rationale for integrated multiomic models.

Methods. We analysed 144 patients with advanced melanoma treated with ICI, divided into a discovery cohort (n=100; 60% with clinical benefit) and a validation cohort withheld from all model development (n=44; 61%). Routine clinical variables were combined with cell-free DNA methylation, measured both as differentially methylated regions and as individual CpG sites, and with serum miRNAs measured at baseline, during treatment, and as change from baseline. Twenty omic combinations across five tiers were evaluated hierarchically using linear predictor offsets, so that each layer was retained only if it improved discrimination in cross-validation beyond established predictors. Overfitting was controlled by selection stability, collinearity pruning and permutation testing. Twelve combinations meeting all prespecified criteria were averaged into a single ensemble score, and a model using clinical variables alone was fitted as a comparator.

Results. The ensemble achieved AUC 0.962 (95% CI 0.917–0.990) in discovery and 0.943 (0.872–0.996) in validation. At a probability threshold of 0.5 it correctly identified 93% of patients with clinical benefit and 90% of those without in discovery (F1 0.93), and 78% and 88% respectively in validation (F1 0.84). Clinical variables alone achieved AUC 0.612 (0.505–0.729) and 0.607 (0.436–0.758), so the omic layers improved discrimination by approximately 0.34 AUC in both cohorts. Methylation carried most of the signal: adding methylated regions to clinical variables raised validation AUC to 0.893, while miRNA change and CpG site methylation reached 0.828 and 0.795 respectively. The signature was robust to simplification: a simple count of how many of the 27 markers fell on their favourable side, dichotomised at a cutoff derived in discovery, achieved validation AUC 0.868.

Conclusion. A multiomic blood test predicted ICI benefit in validation, outperforming clinical variables alone, and may support biomarker-guided treatment selection.