Protein interaction prediction for Alzheimer’s disease using a multi-source protein features fusion framework

• Published in: Informatics and Health Vol. 2; no. 2; pp. 119 - 129
• Main Authors: Yu, Shi-Rui, Yang, Xue-Mei, Sun, Yi-Nan, Li, Yong-Jie, Liu, Yu-Yang, Tang, Xiao-Li
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 01.09.2025; KeAi Communications Co., Ltd
• Subjects: Alzheimer’s disease; Link prediction; Literature; Multi-feature fusion; Protein-protein interaction network; Link prediction; Multi-feature fusion; Protein-protein interaction network; Literature; Alzheimer’s disease
• ISSN: 2949-9534; 2949-9534
• DOI: 10.1016/j.infoh.2025.06.001

Studying the protein interactions associated with Alzheimer’s disease can provide insights into the pathogenesis of this confounding disease. However, to date, researchers have only considered laboratory data and the scientific literature in isolation, and so have not been able to construct a comprehensive protein-protein interaction (PPI) network from which to predict potential interactions. In this study, we devised a framework that integrates protein attributes and interaction information extracted from both experimental data and the scientific literature. Based on these data from multiple sources, We then constructed a PPI network reflecting a diverse range of node features and edge weights. Further, the Graph Convolutional Network (GCN) applicable to the network was used for the link prediction task. Our proposed method achieved superior performance in protein interaction prediction for Alzheimer’s disease with an AUC value of 0.8935. We identified the top ten results predicted by our model as the most promising potential protein interactions for Alzheimer’s disease. The analysis of the predictive results using empirical evaluation verify that our framework has produced a reasonable roadmap for discovering potentially novel protein interactions related to Alzheimer’s disease. •Our method integrated multi-feature PPI data from experimental and literature sources.•A weighted and multi-dimensional PPI network was constructed.•An enhanced GCN model that is able to learn the node features, edge weights, and topological structures of a PPI network was employed to predict novel protein interactions linked to Alzheimer’s disease.•Our method identified top-ten promising unknown protein interactions for Alzheimer’s disease.