Poster Presentation Multi-Omics Conference 2024

Multi-omics Models for Predicting Immunotherapy Resistance in Metastatic Melanoma: Insights from the Personalised Immunotherapy Program (PIP) (#138)

Tuba N. Gide 1 2 3 , Nurudeen A. Adegoke 1 2 3 , Yizhe Mao 1 2 3 , Nigel G. Maher 1 2 3 4 5 , Alison Potter 1 4 5 , Elizabeth C. Paver 1 6 7 , Maria Gonzalez 1 , Andrew J. Spillane 1 2 8 9 , Kerwin F. Shannon 1 2 4 10 , Robyn P.M. Saw 1 2 4 8 , Ismael A. Vergara 1 2 3 , Matteo S. Carlino 1 2 11 , Alexander M. Menzies 1 2 8 9 , Serigne N. Lo 1 2 3 , Ines Pires da Silva 1 2 3 11 , Richard A. Scolyer 1 2 3 4 5 , Georgina V. Long 1 2 3 8 9 , James S. Wilmott 1 2 3
  1. Melanoma Institute Australia, Wollstonecraft, 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. Royal Prince Alfred Hospital, Sydney, NSW, Australia
  5. NSW Health Pathology, Sydney, NSW, Australia
  6. Department of Anatomical Pathology, The Canberra Hospital, Canberra, ACT, Australia
  7. Australian National University, Canberra, ACT, Australia
  8. Mater Hospital, North Sydney, Sydney, NSW, Australia
  9. Royal North Shore Hospital, Sydney, NSW, Australia
  10. Chris O'Brien Lifehouse, Sydney, NSW, Australia
  11. Blacktown and Westmead Hospital, Sydney, NSW, Australia

Introduction: While anti-PD-1±CTLA-4 immune checkpoint inhibitors (ICIs) are effective in treating advanced melanoma, only half of treated patients will survive beyond 5 years, and many experience significant toxicity. The personalised immunotherapy program aims to enhance the prediction of anti-PD-1±CTLA-4 ICI resistance by integrating clinical, molecular, and immunological profiles to identify treatment-resistant melanoma and to prioritise these patients for novel therapies.

 

Methods: This study developed predictive models for ICI resistance in 305 patients with advanced melanoma treated with anti-PD-1±CTLA-4 ICIs, incorporating clinicopathological data, tumour mutation burden (TMB), gene expression profiling (GEP), and spatial quantitative pathology immune profiling (TME). An ensemble of models was used with a consensus nested cross-validation. Models were sequentially developed by adding omics features to clinical factors in a discovery cohort (n=255). Model performance was evaluated using a discriminative index and validated on an independent cohort (n=50).

 

Results: The model developed using baseline clinical factors alone achieved an area under the curves (AUC) of 68% in the discovery cohort. Adding immune cell proportions and spatial interactions from TME to clinical factors increased the AUC to 82%. The addition of TMB and melanoma driver mutations increased the AUC to 83%. GEP-derived immune and stromal signatures increased the AUC to 84% when combined with clinical characteristics. Combining clinical factors, TME, and GEP improved the AUC to 86%. The combination of clinical factors and TMB with either GEP or TME achieved an AUC of 89%. The combination of clinical factors with all omics features yielded the highest AUC (93%). The aggregated risk scores from all omics achieved an AUC of 86% and 92% in the discovery and validation cohorts, respectively.

 

Conclusion: These results underscore the benefit of personalised precision treatment in clinical practice to enhance immunotherapy outcomes in melanoma patients and focus drug development resources to patients most in need.