Identification of metabolite-disease associations based on knowledge graph.

Background: Despite the insights that metabolite analysis can provide into the onset, development, and progression of diseases-thus offering new concepts and methodologies for prevention, diagnosis, and treatment-traditional wet lab experiments are often time-consuming and labor-intensive. Consequently, this study aimed to develop a machine learning model named COM-RAN, which is based on a knowledge graph and random forest algorithm, to identify potential associations between metabolites and diseases.

Methods: Firstly, we integrated the known associations between diseases and metabolites. Secondly, we provided a synthesis of the extant data regarding diseases and metabolites, accompanied by supplementary information pertinent to these entities. Thirdly, knowledge graph-based embedded features were used to characterize disease-metabolite associations. Finally, a random forest algorithm was employed to construct a model for identifying potential disease-metabolite associations.

Results: The experimental results demonstrated that the proposed model achieved an Area Under the Receiver Operating Characteristic Curve (AUC) of 0.968 in 5-fold cross-validations, while the Area Under the Precision-Recall Curve (AUPR) was 0.901, outperforming the vast majority of existing prediction methods. The case studies corroborated the majority of the novel associations identified by COM-RAN, thereby further demonstrating the reliability of the current method in predicting the potential relationship between metabolites and diseases.

Conclusion: The COM-RAN model demonstrated promise in predicting associations between diseases and metabolites, suggesting that integrating knowledge graphs with machine learning methodologies can significantly improve the accuracy and reliability of predictions related to disease-associated metabolites.