HGTDR is a novel approach for drug repurposing that addresses the limitations of existing methods by leveraging heterogeneous graph transformers. The method involves three steps: constructing a heterogeneous knowledge graph, using a heterogeneous graph transformer network, and computing relationship scores with a fully connected network. HGTDR outperforms previous methods in drug repurposing tasks and can predict various relationships, such as drug-protein and disease-protein interactions. The model is end-to-end, scalable, and does not require domain-specific knowledge for setup. It uses PrimeKG as the knowledge graph foundation, combined with BioBERT and ChemBERTa embeddings for node and edge features. The model employs HGT layers for feature extraction and a fully connected network for link prediction. HGTDR is evaluated using 5-fold cross-validation and demonstrates strong performance in predicting drug-disease relationships. The model is robust to input variations and can handle different relation types. It also shows effectiveness in predicting novel drug repurposing candidates, with evidence from medical literature supporting some of these predictions. The method is scalable and can be applied to various tasks beyond drug repurposing, such as predicting other types of relationships in the graph. HGTDR provides a comprehensive solution for drug repurposing by integrating heterogeneous data and utilizing advanced graph neural networks.HGTDR is a novel approach for drug repurposing that addresses the limitations of existing methods by leveraging heterogeneous graph transformers. The method involves three steps: constructing a heterogeneous knowledge graph, using a heterogeneous graph transformer network, and computing relationship scores with a fully connected network. HGTDR outperforms previous methods in drug repurposing tasks and can predict various relationships, such as drug-protein and disease-protein interactions. The model is end-to-end, scalable, and does not require domain-specific knowledge for setup. It uses PrimeKG as the knowledge graph foundation, combined with BioBERT and ChemBERTa embeddings for node and edge features. The model employs HGT layers for feature extraction and a fully connected network for link prediction. HGTDR is evaluated using 5-fold cross-validation and demonstrates strong performance in predicting drug-disease relationships. The model is robust to input variations and can handle different relation types. It also shows effectiveness in predicting novel drug repurposing candidates, with evidence from medical literature supporting some of these predictions. The method is scalable and can be applied to various tasks beyond drug repurposing, such as predicting other types of relationships in the graph. HGTDR provides a comprehensive solution for drug repurposing by integrating heterogeneous data and utilizing advanced graph neural networks.