Construction of knowledge graph of drug-action mechanism

Background: Medical texts contain a large amount of information on the mechanism of action of drug ingredients, and it is impossible to manually collate the massive information in dozens or millions of papers. The method of relationship abstraction in data mining can be used to extract the mechanism of drug action. Methods: First, 770000 papers’ abstracts on the mechanism of the drug in Chinese journals were collected; an algorithm to identify the sentence patterns and semantic elements of the mechanism of drug action were developed; the corresponding semantic elements were extracted by this algorithm. Finally, a knowledge graph was established by the Neo4j tool. Results: A total of 30710 mechanisms of drug action were found, and 57865 drug-action mechanism relations were established. The recall rate of the algorithm to extract semantic elements and their relations was 0.59, the accuracy is 0.86. Conclusion: The algorithm can accurately and comprehensively extract semantic elements and their relationships. Comparison with drug knowledge graphs such as DRKG, this knowledge graph of drug-action mechanism covers a large number of drugs and mechanism of action more comprehensively, which can facilitate new drug development and query of drugs-action mechanism.