How a new knowledge-graph system could reshape drug discovery and personalized cardiology: Why heart scans, genes, and AI are converging to change medicine’s hardest problem
Heart disease has long been medicine’s most stubborn adversary. Despite decades of breakthroughs in imaging, genetics, and pharmaceuticals, cardiovascular disease remains the leading cause of death across Europe, claiming nearly two million lives each year and shaping the daily reality of more than 60 million patients. The paradox is stark: we know more about the heart than ever before, yet translating that knowledge into precise, patient-specific treatments remains painfully slow.
That may be starting to change.
At Imperial College London, researchers have quietly developed an artificial intelligence system that could redefine how heart disease drugs are discovered, tested, and matched to individual patients. The tool, called CardioKG, does not promise miracle cures or instant diagnoses. Instead, it does something far more consequential: it teaches machines to understand how genes, drugs, diseases, and heart function are interconnected, at scale.
If its promise holds, CardioKG signals a shift away from one-size-fits-all cardiology and toward a future where treatments align with how your heart actually works.
Why Heart Disease Still Defies Modern Medicine
Cardiovascular medicine has advanced enormously in diagnostics. We can visualize the heart in extraordinary detail, map electrical signals in real time, and sequence genomes in days. Yet drug discovery remains slow, expensive, and often blunt.
One reason is complexity. Heart disease is not a single condition but a family of disorders, heart failure, atrial fibrillation, coronary artery disease, each shaped by genetics, lifestyle, and physiology. Traditional drug pipelines struggle to account for that diversity. Treatments that work well for one subgroup may fail, or even harm, another.
This is where artificial intelligence, when applied carefully, offers something new: not speed alone, but context.
CardioKG: Teaching AI to Connect the Dots
Unlike conventional AI models that focus on prediction, CardioKG is built around a knowledge graph, a system that links genes, drugs, diseases, and real-world heart imaging into a single, evolving map.
Researchers trained the system using detailed cardiac scans from thousands of participants in the UK Biobank, including patients with heart failure, atrial fibrillation, and prior heart attacks, alongside healthy volunteers. By anchoring genetic and pharmaceutical data to actual heart structure and function, the AI begins to see patterns that siloed datasets often miss.
As Declan O’Regan, who leads the Computational Cardiac Imaging Group at Imperial, explains, the power lies in integration. When imaging data is added to the graph, the system becomes dramatically better at identifying which genes matter, and which drugs might work, for specific cardiac conditions.
This is not theoretical modeling. It is applied insight.
Old Drugs, New Possibilities
One of the most striking findings from the analysis is how it reframes familiar medicines.
The AI highlighted methotrexate, a drug commonly prescribed for rheumatoid arthritis, as a potential candidate for heart failure treatment. It also pointed to gliptins, a class of diabetes medications, as possibly beneficial for patients with atrial fibrillation.
These are not recommendations for off-label use. They are hypotheses, signals emerging from a system capable of seeing connections humans might overlook. Drug repurposing, long considered a shortcut in pharmaceutical development, gains new credibility when guided by integrated biological and imaging data.
Even caffeine appeared in the analysis, with a potential protective association in some atrial fibrillation patients. Researchers were quick to caution against lifestyle changes based on this alone. But the signal itself underscores the system’s sensitivity, and the need for careful interpretation.
From Population Medicine to Personal Hearts
What makes CardioKG particularly significant is not just what it finds, but how it could change care delivery.
Today, most cardiac treatments are optimized for populations. In the future envisioned by the Imperial team, AI systems could help tailor therapies to the functional reality of an individual heart—how it pumps, relaxes, and responds to stress.
Khaled Rjoob, the study’s lead author, describes the next phase as a move toward a “dynamic, patient-centred framework.” Instead of static snapshots, the system would track disease trajectories over time, predicting not only which treatments might help, but when intervention matters most.
This is where AI stops being a research tool and starts becoming a clinical partner.
Beyond the Heart: A Platform, Not a Point Solution
Although CardioKG was developed for cardiovascular disease, its architecture is not heart-specific. The same approach could be applied to brain disorders, metabolic disease, or obesity, any condition where imaging, genetics, and treatment data intersect.
This matters because it reframes AI in healthcare not as a collection of narrow tools, but as an infrastructure for discovery. Instead of chasing isolated breakthroughs, medicine gains a system capable of learning across diseases, populations, and time.
The Caution Beneath the Promise
Yet the excitement should be tempered. AI does not eliminate uncertainty, it redistributes it.
Knowledge graphs are only as good as the data they ingest. Bias in datasets, gaps in representation, and over-reliance on correlations remain real risks. There is also the challenge of translation: insights generated by AI must still survive clinical trials, regulatory scrutiny, and ethical review.
Most importantly, AI must remain interpretable. In life-and-death domains like cardiology, clinicians need to understand why a system suggests a treatment, not simply that it does.
The Imperial team appears acutely aware of these constraints. Their language is cautious, their claims measured. That restraint may be the strongest signal that this work is built for longevity, not headlines.
A Quiet Turning Point
CardioKG will not cure heart disease overnight. But it represents something more durable: a new way of thinking about medicine in an era of overwhelming data.
By teaching machines to understand the relationships between genes, drugs, and how hearts actually function, researchers are laying the groundwork for care that is more precise, more adaptive, and ultimately more human.
In a healthcare system strained by cost, complexity, and chronic disease, that may be one of AI’s most important contributions, not replacing doctors, but giving them clearer maps through the most vital organ we have.
