Imagine a world where discovering drugs for heart disease isn't a decades-long slog through endless trials—now, thanks to cutting-edge AI, we're on the brink of accelerating that process dramatically! But here's where it gets controversial: could this revolutionary tool change how we approach medicine forever, or does it risk overlooking the human element in healthcare?
Knowledge graphs are essentially sophisticated networks that connect dots from vast biological databases. They weave together crucial details about genes, diseases, treatments, molecular pathways, and symptoms into a structured framework, making it easier for researchers to spot patterns. Think of them as a giant, interactive map of biology—until recently, though, these maps missed one vital piece: detailed, personal-level insights into how an affected organ actually appears and operates on an individual basis. For beginners, picture it like having a roadmap of a city that tells you about streets and landmarks but skips the real-time traffic or unique quirks of each neighborhood.
That's where the groundbreaking work from postdoctoral researcher Dr. Khaled Rjoob and group leader Professor Declan O’Regan, both from the Computational Cardiac Imaging Group at the MRC Laboratory of Medical Sciences, steps in. They've enhanced this technology by incorporating imaging data into a knowledge graph for the first time, creating CardioKG—a tool that offers an in-depth look at the heart's structure and function. This addition dramatically boosts the precision of identifying genes tied to diseases and evaluating whether current drugs might offer new treatments. And this is the part most people miss: by visualizing the heart in action, CardioKG turns abstract data into actionable insights, potentially shaving years off drug discovery.
To construct CardioKG, the team drew on heart-imaging data from over 4,280 UK Biobank participants dealing with conditions like atrial fibrillation, heart failure, or heart attacks, alongside 5,304 healthy individuals. This captured a wide array of variations in heart structure and function, generating more than 200,000 image-based traits to train the model. They then fused this with information from 18 varied biological databases and leveraged artificial intelligence to forecast gene-disease links and possibilities for repurposing existing drugs.
As Professor Declan O’Regan explains, 'One of the advantages of knowledge graphs is that they integrate information about genes, drugs, and diseases, giving you more power to uncover new therapies. We discovered that adding heart imaging to the graph completely transformed our ability to pinpoint novel genes and drugs.'
The model pinpointed fresh genes associated with diseases and suggested two drugs for heart conditions: methotrexate, typically used for rheumatoid arthritis, might help with heart failure, and gliptins, prescribed for diabetes, could benefit those with atrial fibrillation. Even more intriguing was their unexpected finding that caffeine, which typically amps up heart excitability, might actually protect patients with atrial fibrillation who experience an irregular, rapid heartbeat. This revelation challenges common assumptions about caffeine's effects—could it be a simple dietary tweak for heart health?
Declan adds, 'What’s exciting is that other recent studies in the field back up our initial results, underscoring the immense promise of knowledge graphs for revealing existing drugs that could be repurposed as new treatments.'
CardioKG serves as a proof-of-concept, proving this approach can stretch beyond the heart. Researchers might now build similar knowledge graphs incorporating imaging data wherever it's available—think brain scans for dementia research, body-fat imaging for obesity studies, or scans of other organs and tissues. This could unlock therapeutic breakthroughs in numerous areas, making drug development faster and more targeted.
Moreover, these graphs can swiftly produce prioritized lists of genes for various diseases, offering pharmaceutical companies a solid launchpad. By spotlighting biological targets to investigate, validate, and develop into therapies, they outpace traditional methods, potentially saving time and resources. For instance, imagine targeting a specific gene linked to heart failure not just through lab tests, but with real imaging evidence guiding the way.
Looking ahead, Dr. Khaled Rjoob notes, 'Building on this work, we'll evolve the knowledge graph into a dynamic, patient-focused system that tracks actual disease progressions. This could pave the way for customized treatments and even predictions of when illnesses might emerge.'
This research received backing from 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.
Beyond his role at the LMS, Declan holds the position of British Heart Foundation Professor of Cardiovascular AI and serves as Clinical Theme Lead for the British Heart Foundation Centre of Research Excellence at Imperial’s National Heart and Lung Institute.
The findings were published in the journal Nature under the title 'A multi-modal vision knowledge graph of cardiovascular disease' on December 29, 2025.
Disclaimer: AAAS and EurekAlert! are not responsible for the accuracy of news releases posted to EurekAlert! by contributing institutions or for the use of any information through the EurekAlert system.
What do you think—should we embrace AI-driven tools like CardioKG to revolutionize drug discovery, even if it means questioning long-held beliefs about substances like caffeine? Do you worry about potential oversights in relying on data over clinical intuition? Share your thoughts in the comments; we'd love to hear if you're excited about personalized medicine or if you see red flags in this approach!
