Purpose/Significance To construct multi-source knowledge graphs, and to propose drug discovery methods based on knowledge graphs, so as to provide reference basis for the research and development of new drugs for diseases. Method/Process The architecture of the knowledge graph is designed, and seven heterogeneous databases are selected. The knowledge graph is constructed through knowledge extraction, data processing and cleaning, and knowledge fusion. Based on the graph and SemMedDB, inference rules and association paths are established, and the weighted link prediction method is improved. A comprehensive knowledge discovery method applicable to drug discovery is proposed. Result/Conclusion Taking Alzheimer’s disease as an example, a multi-source disease knowledge graph containing 67 780 entities and 282 870 triples is constructed, and 184 potential drugs are pvedicted.