Computer-Interpretable Domain Knowledge for Drug-Induced Acute Kidney Injury: a Knowledge Graph Approach.

In this study, a Knowledge Graph (KG) for Drug-Induced Acute Kidney Injury (DAKI) was developed to provide structured and standardized knowledge about drugs known to cause AKI. The DAKI-KG is integrated from several credible domain sources, which we standardized to international vocabularies. We demonstrate the applicability of the DAKI-KG through competency questions and expert evaluations. The envisioned two application areas are: (1) the KG can be used as input for a decision support tool to get insights into potential side-effects and drug-drug interactions harmful for kidneys, especially in patients with chronic kidney disease and (2) researchers can utilize the KG to derive confounders and variables for (causal) machine learning models or as input into link prediction methods. The evaluation results show the applicability of our KG through varying competency questions and use cases, though further improvements are needed before it is ready for clinical application. We anticipate the DAKI-KG to be relevant for tasks such as confounder identification, variable analysis, and the development of QA systems.