From Predicting Cancer Treatment Response to Identifying Novel Therapeutic Targets using Graph Neural Networks.

Abstract

Cancer is a complex disease where treatment resistance frequently arises, posing a significant challenge for patient survival. Predicting the emergence of resistance and understanding its underlying mechanisms are crucial for improving therapeutic strategies. This study introduces TIGENet, an interpretable model for predicting therapy response in oncology and identifying potential therapeutic targets. TIGENet integrates a variational autoencoder for dimensionality reduction with a Graph Neural Network model to predict patient responses to cancer treatments. We employ a graph explainer to highlight the most influential genes driving model predictions, improving interpretability. Our model uses breast cancer transcriptomic and clinical data to identify patients at risk of developing resistance and the key molecular factors involved. By combining tumor- and patient-centered perspectives, these findings contribute to advancing personalized therapeutic interventions.