Two-Channel Graph Inference and Matrix Completion for Predicting CircRNA–Disease Associations in Consumer Health

With the rapid expansion of consumer healthcare, accurately predicting circRNA-disease associations has become essential for advancing disease diagnosis and enabling personalized therapy. However, traditional experimental validation methods are usually costly in terms of both labor and money. In this article, we present an efficient intelligent model, Two-channel Graph Inference based on Global and Local Similarity Networks with Block Matrix Truncation $\gamma $ -norm Minimization (TCGIBMT), aimed at enhancing personalized treatment. First, we integrate multiple similarities between circRNAs and diseases to avoid bias from relying on a single similarity. Second, we introduce a new graph inference technique, GLGI, to handle the sparsity of the association matrix. GLGI captures both global topological insights and local neighborhood details within the circRNA/disease similarity networks, thereby revealing deeper connections while minimizing noise and redundancy from distant nodes. Finally, we propose a novel matrix completion method, BMTNM, to perform the prediction. This method constructs block matrices that encapsulate rich information, substantially reducing computational complexity while retaining robust performance. The truncated $\gamma $ -norm is designed to approximate the matrix rank more effectively by considering both mathematical properties and the matrix’s physical structure. Comprehensive experiments on five datasets show that TCGIBMT consistently outperforms the state-of-the-art model. Our approach’s simplicity, combined with its robust predictive performance, makes it an excellent choice for integration into medical electronic devices aimed at promoting healthier patient habits.