Dual balanced augmented topological noncoding RNA disease triplet association in heterogeneous graphs.

Noncoding RNAs (ncRNAs), including long noncoding RNAs (lncRNAs) and microRNAs (miRNAs), play pivotal roles in various human diseases. Predicting associations such as lncRNA-disease associations (LDAs), miRNA-disease associations (MDAs), and lncRNA-miRNA interactions (LMIs) is crucial for understanding disease mechanisms and identifying therapeutic targets. However, existing models face significant challenges in handling extreme data imbalance and often treat multiple ncRNA-disease and ncRNA-ncRNA interactions collectively, lacking the ability to provide precise, differentiated predictions for specific types of ncRNAs. This limitation reduces their practical applicability. To address these issues, we propose the Dual Balanced Augmented Topological Noncoding RNA Disease triplet Association (DBATNDA) model. DBATNDA constructs an Interaction Dual Graph with LDAs, MDAs, and LMIs as nodes and introduces an efficient graph-based balanced topological augmentation mechanism to enhance node structural representation and adaptability to imbalanced data. This innovative approach enables fast and accurate predictions of ncRNA-disease and ncRNA-ncRNA triplet associations through node classification view. To the best of our knowledge, no existing method employs such a dual-representation strategy to provide simultaneously differentiated predictions for the associations between diverse ncRNAs and diseases while also enhancing target specificity. Experimental results demonstrate DBATNDA's superior performance compared to state-of-the-art models, while case studies confirm its practical significance in these triple association prediction. The code and datasets are publicly available at https://github.com/AI4Bread/DBATNDA.