New AI-driven tool could help find heart disease drugs faster

Knowledge graphs are a powerful tool for bringing together information from biological databases and linking what is already known about genes, diseases, treatments, molecular pathways and symptoms in a structured network. Until now, they have lacked detailed, individual-level information about how the affected organ actually looks and functions.  

The latest research, led by postdoctoral researcher Dr Khaled Rjoob and group leader Professor Declan O’Regan from the Computational Cardiac Imaging Group at the MRC Laboratory of Medical Sciences, has advanced this technology by adding imaging data to a knowledge graph for the first time. CardioKG provides a detailed view of the heart’s structure and function which dramatically improves the accuracy of predicting which genes are linked to disease and whether existing drugs could treat them. 

Capturing heart variation  

To build CardioKG, the team used heart-imaging data from 4,280 UK Biobank participants with atrial fibrillation, heart failure or heart attack, plus 5,304 healthy participants, capturing variation in the structure and function of the heart. In total, over 200,000 image-based traits were generated and used to train the model. The team integrated these with data from 18 diverse biological databases and used artificial intelligence (AI) to predict gene-disease associations and opportunities for drug repurposing.  

“One of the advantages of knowledge graphs is that they integrate information about genes, drugs and diseases,” says Declan, “this means you have more power to make discoveries about new therapies. We found that including heart imaging in the graph transformed how well new genes and drugs could be identified.” 

Predicting new drug opportunities 

The model identified a list of new disease-associated genes and predicted two drugs to treat heart conditions; methotrexate, a rheumatoid arthritis drug, could improve heart failure and gliptins, to treat diabetes, could be beneficial for atrial fibrillation. The team also made a surprising discovery that caffeine, which makes the heart more excitable, has a protective effect in patients with atrial fibrillation who have an irregular and fast pulse. 

“What’s exciting is there are other recent studies in the field which support our preliminary findings,” says Declan, “this highlights the huge potential of knowledge graphs in uncovering existing drugs that might be repurposed as new treatments.” 

Extending the technology to other organs 

CardioKG provides a proof-of-concept technology that can extend far beyond the heart. Researchers could now develop knowledge graphs that integrate imaging data wherever organ imaging exists, meaning the same approach could be applied to brain scans, to body-fat imaging, or to other organs and tissues to explore new therapeutic possibilities in areas such as dementia or obesity. 

The ability of these knowledge graphs to accurately and rapidly generate lists of high-priority genes for a range of diseases would provide pharmaceutical companies with a valuable starting point by highlighting biological targets they can explore, validate and potentially develop into new therapies far more efficiently than traditional discovery methods. 

“Building on this work, we will extend the knowledge graph into a dynamic, patient-centred framework that captures real disease trajectories,” says Khaled, “This will open new possibilities for personalised treatment and predicting when diseases are likely to develop.” 

This study was supported by the Medical Research Council, the British Heart Foundation, Bayer AG, and the National Institute for Health and Care Research (NIHR) Imperial College Biomedical Research Centre.   

Alongside his role at the LMS, Declan is a British Heart Foundation Professor of Cardiovascular AI and Clinical Theme Lead for the British Heart Foundation Centre of Research Excellence within Imperial’s National Heart and Lung Institute. 

---

---