Fig. 1

Analysis workflow for CODEX imaging data. A CODEX cyclical staining was performed on an initial cohort of 42 ICI treated patients. Significant differences in overall survival were evident between patient response groups B Images were imported into Qupath and segmented with Cellpose. A pixel classifier was trained on cytokeratin to create a tumour mask. This mask was expanded to capture cells at the tumour periphery. C Images were visually inspected for QC, and six cores that were fragmented, necrotic or contained imaging artifacts were excluded. Response information was not available for one patient. D Marker QC was performed and markers with non specific signal, or artifactual staining were excluded. E Cell phenotyping was performed in python by unsupervised clustering and cell annotation by canonical marker expression. F Spatial metrics were generated, including cell densities, pairwise cell–cell interaction, neighbourhood detection and cell type enrichment within these neighbourhoods, as well as published SpatialScore metric. G These spatial metrics were analysed for differences between ICI response groups and patient survival times