Multi-Omics Data-Based Breast Cancer Relapse Prediction Using Optimized Graph Neural Network

In oncology, breast cancer relapse remains a significant challenge due to the lack of accurate and personalized predictive models for timely clinical intervention. With this motivation, this study proposes a multi-omics optimized graph neural network (GNN) for relapse prediction based on gene expression, proteomics, and clinical features. The proposed method creates a patient-patient graph using cosine similarity and utilizes Laplacian-normalized graph convolution layers with dropout regularization and L2 weight decay. The Adam optimizer is used to optimize the training. Evaluation metrics justify the proposed approach regarding accuracy, precision, recall, F1–score, and AUC-ROC. On an artificially created dataset of 500 patients, experimental results demonstrate that the proposed model outperforms traditional machine learning baselines and gives much promise for actual world application in personalized medicine and clinical decision support systems.