FHG-GAN: Fuzzy Hypergraph Generative Adversarial Network With Large Foundation Models for Alzheimer’s Disease Risk Prediction

Risk prediction for Alzheimer's Disease (AD) is crucial for delaying disease progression in clinics. However, existing methods face significant challenges in leveraging the high-order and fuzzy associations in biomedical data. To address this limitation, we propose a fuzzy hypergraph-based deep learning framework to explore multi-omics associations and enhance predictive performances efficiently. In specific, we first establish a mathematical model of fuzzy structural entropy propagation. This model represents multi-omics associations by fuzzy hypergraphs, characterizing the progression of AD as the topological evolution of fuzzy hypergraphs. Next, we design a Fuzzy Hypergraph Generative Adversarial Network (FHG-GAN). Particularly, large foundation models are used to generate high-quality feature representations, alleviating data noise and inconsistencies. FHG-GAN captures the evolutionary patterns of diseases using fuzzy hypergraph convolutional layers designed based on the mathematical model, thereby achieving accurate disease risk prediction and pathogeny extraction. Finally, we experimentally demonstrate the superiority of FHG-GAN compared with advanced methods, indicating FHG-GAN's ability to support the early diagnosis and treatment of AD-like diseases. Our code is published on: github.com/fmri123456/FHG-GAN.