Drug repurposing is a cost-effective approach for discovering new indications for existing drugs. In current research, leveraging multi-omics biological information through artificial intelligence methods for drug repurposing has become a hot topic. The MetaGT-HGN proposed in this paper is constructed on the basis of a heterogeneous graph containing entities such as drugs, diseases, genes, and targets. It fully integrates multi-omics biological information and utilizes the powerful structural information extraction ability of the graph transformer module. Meta learning can provide excellent generalization ability for multidrug-disease association prediction tasks. The MetaGT-HGN outperforms existing methods in drug repurposing tasks and yields better results in the task of discovering new indications for existing drugs. Furthermore, the reliability and utility of the prediction results of this paper are verified, which provides significant assistance for the rapid development of new treatment methods.