Poster Presentation Multi-Omics Conference 2024

Explainable machine learning identifies features and thresholds predictive of immunotherapy response (#137)

Venkateswar Addala 1 , Khoa A Tran 1 2 , Lambros T Koufariotis 1 , Jia Zhang 1 , Scott Wood 1 , Conrad Leonard 1 , Lotte L Hoeijmakers 3 , Christian U Blank 3 4 , Mireia Crispín Ortuzar 5 6 , Elizabeth D Williams 2 7 8 , Olga Kondrashova 1 2 , John V Pearson 1 , Nic Waddell 1
  1. Cancer Program, QIMR Berghofer Medical Research Institute, Brisbane, QLD, Australia
  2. School of Biomedical Sciences, Queensland University of Technology (QUT), Brisbane, QLD, Australia
  3. Department of Medical Oncology, Netherlands Cancer Institute (NKI), Amsterdam, The Netherlands
  4. Department of Medical Oncology, Leiden University Medical Center, Leiden, The Netherlands
  5. Department of Oncology, University of Cambridge, Cambridge, UK
  6. Cancer Research UK Cambridge Centre, University of Cambridge, Cambridge, UK
  7. Australian Prostate Cancer Research Centre, Queensland (APCRC-Q) and Queensland Bladder Cancer Initiative (QBCI), Brisbane, QLD, Australia
  8. Centre for Genomics and Personalised Health, Queensland University of Technology (QUT), Brisbane, QLD, Australia

Cancer immunotherapy has improved patient survival for multiple cancer types, including melanoma. A variety of molecular features have been linked to response to immune checkpoint inhibitors (ICI) treatment1-3. Clinically established biomarkers, tumour mutation burden (TMB)4 and expression of PD-L15, have not effectively fulfilled the role of accurately categorising responders versus non-responders. Due to the complex nature of ICI response, which includes cancer intrinsic features and cancer extrinsic features within the tumour microenvironment (TME)6, using a single biomarker to predict response is not sufficient and necessitating a need to identify accurate clinical and molecular predictors of treatment response. Here we integrate clinical, DNA and RNA sequencing data from four datasets, comprising 138 melanoma patients treated with ICI, to develop machine learning models of ICI response. We evaluated the performance of each model using an independent dataset of patients with cutaneous melanoma (n=53). The best predictive model was the multi-omic ensemble model combining logistic regression and random forest, with AUC-ROC of 0.78 when predicting response in the independent cutaneous dataset. Using the explainability method SHAP we predicted thresholds for mutational signatures, neoantigen load, immune cell-type abundance and immune receptor LAG3 expression as key features linked to response. These thresholds may be specific to sample processing and analytical approaches employed. The model was applied to additional non-cutaneous melanoma patients (n=15) correctly predicting response for 11 cases. This approach highlights patient response to ICI involves cancer intrinsic and extrinsic features and identifies candidate biomarkers to inform the use of ICI in melanoma.

  1. D. Liu et al., Integrative molecular and clinical modeling of clinical outcomes to PD1 blockade in patients with metastatic melanoma. Nature Medicine 25, 1916-1927 (2019).
  2. F. Newell et al., Multiomic profiling of checkpoint inhibitor-treated melanoma: Identifying predictors of response and resistance, and markers of biological discordance. Cancer Cell 40, 88-102 e107 (2022).
  3. E. M. Van Allen et al., Genomic correlates of response to CTLA-4 blockade in metastatic melanoma. Science 350, 207 (2015).
  4. L. Marcus et al., FDA Approval Summary: Pembrolizumab for the Treatment of Tumor Mutational Burden-High Solid Tumors. Clin Cancer Res 27, 4685-4689 (2021)
  5. A. Fusi et al., PD-L1 expression as a potential predictive biomarker. Lancet Oncol 16, 1285-1287 (2015)
  6. V. Addala et al., Computational immunogenomic approaches to predict response to cancer immunotherapies. Nat Rev Clin Oncol, (2023).