<p>Biomedical knowledge graphs (KGs) such as Hetionet provide rich multi-relational structures connecting diverse biological and clinical entities, yet learning effective representations remains challenging due to heterogeneity, semantic complexity, and data incompleteness. We propose MVCL, a biomedical knowledge graph–oriented Multi-View Contrastive Learning framework that explicitly aligns and fuses heterogeneous structural, textual, and ontological views through a unified contrastive objective. Unlike generic multi-view contrastive approaches, MVCL performs cross-view node-level contrastive alignment to enforce semantic consistency among graph structure, biomedical text, and ontology hierarchies. Specifically, MVCL integrates (i) a structural view modeled by a Heterogeneous Attention Network over biologically meaningful meta-paths, (ii) a semantic textual view encoded using BioBERT, and (iii) an ontology hierarchy view learned via a graph neural network over curated biomedical ontologies such as gene ontology (GO) and Medical Subject Headings (MeSH). Extensive experiments on Hetionet show that MVCL consistently outperforms strong baselines across link prediction, node classification, and clustering, achieving up to 2.3% higher node classification accuracy, 2.8% improvement in link prediction Area Under the ROC Curve (AUC), and 8% gain in clustering Normalized Mutual Information (NMI). Due to the simultaneous optimization of multiple encoders and large-batch contrastive learning, all experiments are conducted on multi-GPU high-performance computing infrastructure, highlighting MVCL’s alignment with supercomputing environments.</p>