Associate Professor, University of Queensland 

Immune checkpoint inhibitors (ICI) have improved clinical outcomes for patients with advanced non-small cell lung carcinoma (NSCLC), however a subset of patients remain treatment resistant. Spatial biology offers key intelligence into cellular coordinates in situ, with accurate interpretation requiring both biological and computational consideration. We analysed >600 NSCLC biopsies taken from patients prior to ICI treatment and applied a deep-learning model to classify cells into distinct phenotypes, map spatial regions for tissue compartments and metabolic neighbourhoods, and perform statistical comparisons for the measured features in these regions for clinical benefit at 6 months following ICI treatment. Geometric profiling of spatial interactions at a range of scales was accomplished through feature engineering, followed by statistically robust stability selection of features. Modelling of the patient cohort relapse events by survival analysis demonstrated that these features largely recapitulate a role for metabolic and dynamics of the tumour microenvironment and predict patient response to ICI.