Mobile Menu

Using a knowledge graph to discover adverse reactions

Researchers have constructed a tumour-biomarker knowledge graph and used it to discover potential adverse drug reactions from antitumour drugs.

Adverse drug reactions

Adverse drug reactions (ADRs) are the cause of significant morbidity and mortality in patients. It is also a source of financial burden across healthcare systems. Patients with tumours experience relatively high rates of ADRs from antitumour drugs. They also more easily experience rare and severe ADRs, which can seriously impact their quality of life. The identification of these ADRs during the premarket period is limited by the small sample size and generalisability of clinical trials. Exploring these reactions is critical to decrease the incidence. Therefore, there has been a lot of effort in detecting potential ADRs through data mining of literature databases or electronic health records.

The authors in this paper define knowledge graphs as data models that represent facts as nodes and relations between the nodes. In medical information networks, objects such as diseases, drugs, biomarkers or treatment can all be linked together. This enables the discovery of knowledge on a scale and at a speed that traditional pharmacologic experiments or clinical trials cannot approach.

Researchers are increasingly adopting biomarkers to accurately and specifically predict therapeutic efficacy and safety during cancer therapy. However, there is little effort applying this to assess ADRs.

Tumour-biomarker knowledge graph

In this study, published in Frontier in Genetics, researchers aimed to discover potential ADRs from antitumour drugs and provide explanations by constructing a knowledge graph using literature data sources.

Using machine-learning methods, the team constructed a tumour-biomarker knowledge graph using biomedical literature. This graph contains four types of node: tumour, biomarker, drug and ADR. Using this model, the team found potential ADRs of antitumour drugs and also provided explanations. Experiments on real-world data showed that the model could achieve 0.81 accuracy. Their model also outperformed traditional co-occurrence methods. Each calculated ADR was attached with the corresponding paths of ‘tumour-biomarker-drug’. Capturing this detailed information can help obtain in-depth insights into the underlying mechanisms.

This tumour-biomarker knowledge-graph based approach is an explainable method for potential ADR discovery based on biomarkers. This may be a valuable tool to those working on biomedical literature mining. It may also catalyse future research into the mechanism of ADRs as well as identification of biomarkers to predict ADRs.

Image credit: Nastya Dulhiier – Unsplash

Share this article