Jihwan Ha
Long non-coding RNAs (lncRNAs) are a diverse class of transcripts that regulate gene expression through chromatin remodeling, transcriptional control, and RNA processing. Their roles have been increasingly implicated in various biological processes and diseases, including development, immunity, cancer, and neurological disorders. Elucidating the molecular pathways by which lncRNAs contribute to disease pathology offers significant potential for the development of innovative diagnostic biomarkers and targeted therapeutic interventions. Given the limitations of traditional wet-lab methods-particularly in terms of scalability, cost, and labor-computational frameworks have emerged as indispensable tools for systematically analyzing lncRNA-disease associations and guiding hypothesis-driven experimental research. In this study, we propose TBLDA, a novel transfer learning-based framework for lncRNA-disease association prediction. TBLDA leverages BERT-derived embeddings to capture rich semantic representations of disease terms and integrates a pre-trained miRNA-disease association model to transfer task-related knowledge, thereby enhancing predictive performance in the lncRNA-disease prediction task. In conclusion, TBLDA demonstrated robust predictive performance, attaining AUC scores of 0.9801 and 0.9721 under leave-one-out cross-validation (LOOCV) and 5-fold cross-validation (5-fold CV), respectively. These results consistently outperformed five competitive baseline models, highlighting the effectiveness of our transfer learning strategy. In summary, the proposed framework offers a powerful computational tool that can complement experimental research and contribute to the identification of disease-associated lncRNAs, supporting further studies in medical research.