Search for a command to run...
Accumulating evidence demonstrates that microRNA (miRNA) dysregulation drives the pathogenesis of diverse human diseases via intricate, context-dependent molecular mechanisms. Hence, prediction of miRNA-disease association types is a critical prerequisite for dissecting functional roles of miRNAs in disease initiation and progression. Although computational methods offer cost-effective, time-efficient alternatives to wet-lab experiments for miRNA-disease association type prediction, most of them are hampered by three key limitations: excessive reliance on association-derived similarity metrics gives rise to quantification bias, traditional pairwise graph architectures inadequately capture high-order biological interactions, and existing representation learning strategies fail to generate consistent embeddings across heterogeneous views and modalities. To address these issues, this study presents DHGCMDA, a dual-view heterogeneous graph contrastive learning framework for miRNA-disease association type prediction. Specifically, dual-view hypergraphs are first constructed based on heterogeneous similarity data to avoid excessive reliance on association-derived similarity metrics. A hypergraph convolutional network is then employed to capture high-order topological relationships between miRNAs and diseases, with its convolution cooperatively integrated with contrastive learning, intra-modality for cross-view consistency and cross-modality for embedding space alignment, to enhance feature representation quality. Finally, an attention-guided adaptive view fusion strategy dynamically weights and integrates distinct view representations, and type-aware message passing via heterogeneous graph Transformer simultaneously enables prediction of association presence and functional types. 5-fold cross-validation on HMDD v2.0 and v3.2 datasets demonstrates that DHGCMDA outperforms several state-of-the-art methods. Furthermore, case studies on breast neoplasms and hepatocellular carcinoma reveal that most predicted association types are corroborated by published literature, thereby validating the efficacy of DHGCMDA in miRNA-disease association type prediction. DHGCMDA exhibits robust discriminative power and generalization capability, providing a reliable computational alternative for miRNA-disease association type prediction. The source code is publicly available at https://github.com/CDMBlab/DHGCMDA.