Multi-view graph neural network with channel attention and learnable view-weight fusion for lncRNA-disease association prediction.

Accumulating evidence suggests that long noncoding RNAs (lncRNAs) interact synergistically with diverse cellular molecules to regulate a wide range of biological processes. Identifying lncRNA-disease associations (LDAs) is therefore essential for elucidating disease mechanisms and advancing biomarker discovery. Although recent computational methods have incorporated heterogeneous biological entities and complex graph structures, the effective preservation of view-specific information and the adaptive integration of complementary biological features remain insufficiently explored. This study propose MVMCLDA, a multi-view graph neural network with a channel attention module and learnable view-weight fusion. Experimental results demonstrate that MVMCLDA achieves an area under the ROC curve (AUC) of 0.9876 ± 0.0147 and an area under the precision-recall curve (AUPRC) of 0.9874 ± 0.0154 under 5-fold cross-validation, outperforming the compared baseline methods and demonstrating robust predictive performance. Ablation experiments further confirm the effectiveness of the multi-view feature integration strategy and the proposed model architecture. In addition, case studies on gastric cancer, ovarian cancer, and breast cancer support the biological plausibility of the predicted associations.