Scientists have developed an artificial intelligence (AI) system that can assess an individual's risk of venous thrombosis with greater accuracy by integrating clinical information, genetic data, and patterns of gene activity.
The study, published in the Journal of Thrombosis and Haemostasis, identified hundreds of previously unrecognized molecular markers associated with the disease, bringing researchers a step closer to personalized strategies for thrombosis prevention.
Venous thrombosis remains one of the most common cardiovascular disorders. A particularly challenging form is idiopathic venous thromboembolism (VTE), which occurs without an obvious triggering factor and is therefore difficult to predict before the first thrombotic event.
Researchers from the Sant Pau Research Institute and the Center for Biomedical Network Research on Rare Diseases (CIBERER) analyzed data from 790 individuals belonging to families with a history of venous thromboembolism.
Among the participants were 70 individuals who had previously experienced idiopathic venous thrombosis. The study was conducted as part of the GAIT2 (Genetic Analysis of Idiopathic Thrombophilia) project, one of the largest research programs investigating inherited susceptibility to thrombosis.
Unlike conventional risk assessment methods, the researchers combined multiple layers of biological information. In addition to clinical characteristics and genetic variants, they analyzed the activity of 12,981 genes—the transcriptome, representing all genes actively expressed in the body's cells.
Machine learning algorithms simultaneously processed thousands of biological variables to uncover hidden patterns associated with thrombosis risk.
The study's most significant finding was the identification of 494 genes whose activity could reliably distinguish individuals who had experienced thrombosis from those who had never developed the condition.
Notably, many of these markers were long non-coding RNAs (lncRNAs)—regulatory RNA molecules whose role in venous thrombosis has remained largely unexplored.
Using these data, the researchers created a molecular "thrombosis signature." The AI model generates a similarity score that indicates how closely an individual's molecular profile resembles that of patients who have already experienced a thrombotic event.
This approach enabled the identification of people with no clinical signs of disease but whose molecular profiles closely matched those of high-risk patients.
Adding transcriptomic data substantially improved the model's predictive accuracy. When only clinical and genetic information was used, 43% of individuals who had never experienced thrombosis were incorrectly classified as high risk. After incorporating gene expression data, the false-positive rate fell to 23%.
At the same time, the model became better at identifying individuals who had truly experienced thrombosis, with detection accuracy increasing from 70% to 74%.
The study also confirmed associations between venous thrombosis and several biological pathways that had previously been suspected to play a role in the disease. These included molecular mechanisms linked to heart muscle disorders and the function of the proximal renal tubules, further supporting the biological relevance of the newly identified molecular signatures.
Follow NEWS.am Medicine on Facebook and Twitter