GraphX-Net: A Graph Neural Network-Based Shapley Values for Predicting Breast Cancer Occurrence

Abstract

Breast cancer is a major health problem worldwide, and an accurate prediction of its recurrence is crucial to early detection of recurrence and personalized treatment. In recent years, various AI techniques have been applied to predict cancer recurrence with increasingly high accuracy. Graph Neural Networks (GNNs) have emerged as powerful tools for analyzing structured data, including knowledge networks. In this study, we explore the application of GNN-based node classification to predict the occurrence of relapse status in breast cancer patients. We propose a noval model, namely GraphX-Net, Shapley Value-based Graph Neural Network. GraphX-Net leverages a graph representation by analyzing public breast cancer gene expression data, where nodes represent patients and edges capture relationships between them based on various factors such as tumor cellularity, histological subtype, hormone therapy, patient vital status, primary tumor laterality, type of breast surgery, and other clinicopathological parameters. Our approach integrates diverse data sources and utilizes graph convolutional layers to generate informative node embeddings. The model uniquely calculates node feature contributions using Shapley values, sets nodes’ thresholds, and considers the total node neighboring effects. These innovations enable GraphX-Net to achieve state-of-the-art performance in predicting breast cancer recurrence, highlighting its potential as a powerful tool for improving patient outcomes.

Keywords

• Graph neural networks
• XAI
• node classification
• breast cancer
• occurrence prediction