Multiomics-based graph convolutional neural network for Alzheimer’s disease diagnosis and MCI progression prediction

Abstract

Alzheimer’s disease is a degenerative disorder of the central nervous system. In the early stage, its symptoms can be easily confused with those of aging, causing patients to miss the optimal treatment period and exacerbating the condition. This study aims to assist doctors in diagnosing Alzheimer’s disease and predicting the progression of mild cognitive impairment by using graph convolutional neural networks. The CA-GCN model was constructed by building an adjacency matrix based on sample similarity, introducing Chebyshev convolution for hierarchical analysis, and leveraging an adaptive mechanism based on clinical information to flexibly adjust the relationships among samples. The experimental results demonstrate that the CA-GCN model has a stable accuracy in diagnosis (AUC, 0.97) and prediction experiments (AUC, 0.88), outperforming common machine learning algorithms. This model improves diagnostic accuracy and assists clinical decision-making, predict disease progression, and thus treat patients in a timely manner, reducing the burden on families.