The Future of Cardiogenetics Lies in Artificial Intelligence

© iStock

Researchers from the AI and Digital Science Institute at the HSE Faculty of Computer Science have developed a program capable of analysing regions of the human genome that were previously inaccessible for accurate interpretation in genetic testing. The program adapts large generative AI (GenAI) models for cardiogenetics to predict how specific mutations affect the function of individual genes.

The human genome can be likened to an enormous library. Until recently, scientists could read only a small fraction of its 'books'—those that contain instructions for making proteins (about 2% of the total DNA). This is where researchers typically looked for mutations responsible for hereditary heart diseases. For such variants, well-established international criteria exist to assess their risk and identify mutations that drive disease progression. But what about the remaining 98%? For a long time, these regions were dismissed as 'empty pages' or 'genetic junk.' However, it has become clear that they are far from useless: they function as switches and volume controls, regulating how actively genes are expressed. Disruptions in these regions can significantly affect the functioning of the heart, blood vessels, and blood. 

The challenge was that scientists were previously unable to determine which of these 'invisible' mutations were truly harmful and which were benign. As a result, many cases of heart disease remained unexplained due to the inability to analyse non-coding variants, ie those that do not contain direct instructions for protein synthesis. Researchers at the HSE FCS AI and Digital Science Institute have proposed a software solution that, for the first time, enables large-scale, accurate analysis of these 'silent' regions in the context of heart health. The software leverages state-of-the-art generative models—the technology underlying popular neural networks—to predict the effects of mutations in regulatory DNA regions and assess their impact on cardiovascular function.

Maria Poptsova, Director of the Centre for Biomedical Research and Technologies at the HSE FCS AI and Digital Science Institute

'The program is built on two powerful AI models acting as experts who have read millions of genetic instructions and are therefore able to compare two DNA variants: a healthy, or reference, sequence and a sequence in which a mutation has occurred. The program then assesses whether the "volume" of the genes has changed as a result of the mutation, meaning whether they have become more active or, conversely, less active. We focused on heart and blood vessel tissues, but the method can be applied to any tissue.'

To improve accuracy, the program uses a form of collective intelligence: several models analyse each mutation from different perspectives, and their findings are then integrated using artificial intelligence methods. As a result, the program produces a simple, interpretable score between 0 and 1. The closer the score is to 1, the higher the likelihood that the detected mutation is harmful and may contribute to the development of heart disease.

To ensure the program is reliable, the scientists conducted a rigorous validation study. They used data from the UK Biobank project, a large-scale database of genetic information. For testing, more than 11,000 mutations were selected from the regulatory regions of DNA that had previously been difficult to analyse. The dataset included both variants already known to be associated with disease and clearly benign variants. To ensure a fair experiment, each potentially harmful mutation was compared with nine benign ones selected based on the maximum number of matching characteristics: genomic location, site type, proximity to genes, and other parameters. The program successfully completed the task, reliably distinguishing pathogenic mutations from harmless ones and demonstrating its robustness and readiness for practical application.

The program was developed for practical use by a wide range of specialists, including staff in medical laboratories and cardiology centres, who will be able to interpret genome-wide sequencing results more accurately and identify genetic causes of disease in patients. It is already being introduced into the workflows of genetic laboratories. As the developers note, no programming skills are required to use the system: it is designed for everyday use by geneticists, bioinformaticians, and medical researchers. 

In basic research, the program can help understand the molecular mechanisms underlying the development of heart disease and explore how regulatory DNA regions contribute to pathology. Using this tool, scientists at the HSE FCS Centre for Biomedical Research and Technologies have already made an important discovery: certain variants of the BMPR2 gene that affect its activity can influence how a patient responds to treatment. The researchers are now continuing their work, focusing on non-coding DNA regions that affect the function of genes associated with the risk of sudden cardiac death. 

The GenAI model 'Predicting the Effect of Non-Coding Variants Based on the Adaptation of GenAI Models to the Cardiogenetics Domain' was developed as part of a programme implemented by the HSE AI Research Centre under a grant from the Russian Ministry of Economic Development.