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

We present a Knowledge Graph (KG) for Drug-Induced Acute Kidney Injury (DAKI). The DAKI-KG contains domain knowledge about drugs known to cause acute kidney injury (AKI), such as their indications, contra-indications, adverse events and drug-drug interactions. The aim of the DAKI-KG is to be applied to a varying number of tasks, such as variable (drug) analysis, confounder analysis and research surrounding QA systems.

We developed our KG through a semi-automated process with a strong focus on trustworthy domain knowledge sources and expert knowledge. Through the automated processes, it is possible that some facts are not fully complete or accurate, hence we do not recommend the DAKI-KG to be used in the clinical practice. In our paper, we describe the full integration and evaluation process.

This work was conducted as part of the LEAPfROG project supported by Nederlandse Organisatie voor Wetenschappelijk Onderzoek (NWO; Dutch Research Council) (KICH1.ST01.20.011) and co-funded in cash by Dutch Kidney Foundation and National Intensive Care Evaluation (NICE) foundation, and in kind by PHARMO Institute for Drug Outcomes Research, Castor, InsightRX, Z-Index, Digital Health Link.

Here we deposit the files that form the the DAKI-KG (ttl/ttls format): 

The ontology file describes the classes and properties defined within the dakikg namespace.

The intrinsic_drug_properties file contains the indications, contra-indications and side-effects.

The drug_interactions file contains the drug-drug interactions

The aki_risk_factors file contains drugs and concepts that are AKI risk factors

The nephrotoxic_classes file contains the concepts that fall under the AKI or CKD class.

We also have files containing solely metadata:

elt_metadata, containing all triples derived by the ELT, the similarity score from the ELT, the isolated fragment the ELT matched upon, and the extraction date,

drug_inclusion_rationales, describing the reason for inclusion for the main drugs of focus (i.e., through being an indication for CKD or one of its main comorbidities).

We also recommend including the snomed_ct_hierarchy and snomed_ct_pt files, which represent the SNOMED CT hierarchy and preferred terms: these are not publicly available through the repository due to to licensing restrictions. The DAKI-KG can be used without those files, but we strongly recommend to include the SNOMED CT hierarchy file (resulting in the {sct:Concept rdfs:subClassOf sct:Concept} triples) to retrieve the benefits of hierarchical querying. Code and instructions for this can be found on our GitHub repository. 

Our GitHub repository also contains other files related to the DAKI-KG, such as relevant SPARQL queries: GitHub Repo (can also be retrieved through our permanent URL).The code used to integrate the DAKI-KG can be found on our code repository.