Immune checkpoint inhibitors (ICIs), which reactivate T-cell responses against tumours, show limited clinical efficacy due to low response rates and the lack of robust predictive biomarkers. Cell–cell interactions within the tumour microenvironment influence therapeutic response and can be analysed at single-cell resolution using imaging-based spatial transcriptomics. We present GraphTME, a spatially informed and biologically interpretable framework that predicts anti-PD-1 response by modelling pathway-specific ligand-receptor signalling as a multi-relational directed graph with edge weights inversely scaled by spatial distance. Using CD8+ T cells from single-cell RNA sequencing data of ICI-treated non-small cell lung cancer (NSCLC) patients, we trained a model to infer immune responsiveness. GraphTME achieved an F1 score exceeding 0.83 in predicting ICI response and was examined using MERFISH data from NSCLC patients with clinical responses. In these patients, CD8+ T cells predicted as responders exhibited higher abundance and prominent directional signalling towards tumour cells. These cells also expressed genes associated with antitumour activity. GraphTME is among the first frameworks to quantitatively capture single-cell-level interactions within spatial tumour architecture and leverage them for ICI response prediction. It offers a spatially resolved, biologically grounded biomarker for immunotherapy and a tool for dissecting immune dynamics in situ.
Jeong et al. (2026) studied this question.