A Novel Kernel-Based Hilbert Space Framework for Predictive Modeling of lncRNA-miRNA-Disease Interaction Networks

Long non-coding RNAs (lncRNAs) and microRNAs (miRNAs) are vital regulators of gene expression. They are closely linked to the development and progression of various diseases, including cancer and neurodegenerative disorders. Their interactions form complex lncRNA-miRNA-disease networks (LMDNets), which modulate key disease-associated biological pathways. Although recent computational approaches have attempted to integrate diverse biological data such as gene expression profiles, sequence information, and disease associations. They often face significant challenges, including limited interpretability, reliance on manually curated datasets, poor scalability, and high sensitivity to noise and missing data. To overcome these limitations, we propose KHSF-LMDNet, a kernel-based Hilbert space framework for modeling lncRNA-miRNA-disease networks. This model integrates graph-based biological networks, similarity features, and deep learning with an attention mechanism to map complex interactions into a Hilbert subspace, enabling more robust and interpretable learning. Evaluated on benchmark datasets, KHSF-LMDNet consistently outperforms existing methods in terms of accuracy, precision, and area under the ROC curve (AUC). Additionally, it effectively ranks disease-associated lncRNAs and miRNAs, identifying top candidates related to cancer and Alzheimer's disease, thereby supporting functional genomics research and facilitating the discovery of novel biomarkers for precision medicine.