Biological knowledge graph-guided investigation of immune therapy response in cancer with graph neural network.

This article focuses on the development and evaluation of ICInet, a deep learning framework guided by multiple prior biological knowledge networks, designed to predict patient responses to immune checkpoint inhibitors (ICIs) in cancer therapy. ICInet integrates gene expression profiles with gene regulatory and protein interaction networks using graph convolutional neural networks to improve prediction accuracy across multiple cancer types, including melanoma, gastric cancer, and bladder cancer. Tested on over 600 ICI-treated patient samples from seven cohorts, ICInet consistently outperformed traditional biomarkers such as tumor mutational burden and PD-L1 expression, demonstrating robust generalizability across datasets and cancer subtypes. The study highlights ICInet’s potential to identify immunotherapy-response-associated biomarkers, thereby advancing precision oncology by enabling more accurate prediction of immunotherapy outcomes.