GraphTME: graph-based framework for predicting immunotherapy response by interpreting tumour microenvironment...

Area of Science:

• Oncology
• Immunology
• Computational Biology

Background:

• Immune checkpoint inhibitors (ICIs) show limited efficacy due to low response rates and lack of predictive biomarkers.
• Tumor microenvironment cell-cell interactions influence therapeutic response and can be studied using spatial transcriptomics.

Purpose of the Study:

• To develop a spatially informed and biologically interpretable framework (GraphTME) for predicting anti-PD-1 response.
• To leverage single-cell RNA sequencing and MERFISH data to model ligand-receptor signaling for immune responsiveness prediction.

Main Methods:

• Modeled pathway-specific ligand-receptor signaling as a multi-relational directed graph with distance-scaled edge weights.
• Trained a model using CD8+ T cells from ICI-treated non-small cell lung cancer (NSCLC) patients.
• Validated the framework using MERFISH data from NSCLC patients.

Main Results:

• GraphTME achieved an F1 score exceeding 0.83 in predicting ICI response.
• Predicted responders showed higher CD8+ T cell abundance and directional signaling towards tumor cells.
• Responder CD8+ T cells expressed genes linked to antitumour activity.

Conclusions:

• GraphTME quantitatively captures single-cell interactions within tumor architecture for ICI response prediction.
• The framework provides a spatially resolved, biologically grounded biomarker for immunotherapy.
• GraphTME serves as a tool for dissecting in situ immune dynamics.