PathTIGR: A pathway topology-informed graph representation learning framework for immunotherapy response prediction.

Area of Science:

• Computational Biology
• Genomics
• Cancer Research

Background:

• Immunotherapy has transformed cancer treatment but faces challenges due to patient response variability.
• Predicting immunotherapy efficacy remains difficult, limiting its benefit to certain patient groups.

Purpose of the Study:

• To develop PathTIGR, a graph representation learning framework for predicting immunotherapy response.
• To integrate biological pathway network topology with genomic variation data for enhanced prediction accuracy.

Main Methods:

• PathTIGR employs a three-component architecture: a pathway graph encoder with multihead attention, a transformer module for regulatory dependencies, and a multilayer perceptron for prediction.
• The framework embeds pathway topology and cancer genomic variants into pathway representations.

Main Results:

• PathTIGR demonstrates superior predictive performance compared to existing biomarkers and deep learning methods across multiple cohorts.
• The framework provides biological interpretability by identifying key signatures associated with response heterogeneity.

Conclusions:

• PathTIGR offers an interpretable, graph-based learning approach to improve immunotherapy response prediction.
• This framework elucidates molecular determinants of therapeutic efficacy, advancing precision cancer immunotherapy.