AI Model Boosts Personalized Medicine for Genetic Diseases

Cedars-Sinai investigators have developed a novel artificial intelligence (AI) model, named DYNA, that accurately distinguishes harmful gene variations from harmless ones, potentially enhancing physicians’ ability to diagnose diseases. The new tool could pave the way for more precise personalized medicine and targeted therapies.

Published in the peer-reviewed journal Nature Machine Intelligence, the team demonstrated that DYNA outperforms existing AI models in accurately predicting which changes in DNA, commonly called mutations, are linked to specific cardiovascular conditions and other disorders.

“In recent years, AI has vastly expanded our ability to detect enormous numbers of genetic variants in ever-larger populations,” said Huixin Zhan, PhD, a contributing author of the study from the Department of Computational Biomedicine at Cedars-Sinai. “But up to half of these variants are of uncertain significance, meaning we don’t know if they cause a disease and, if so, which one. The DYNA model overcomes many of these challenges.”

Current AI models prove to be effective in distinguishing which gene variants, in general, are more likely to negatively affect the structure or function of a protein, which can lead to disease, Zhan said. But these models lack the ability to connect a specific variant to a specific disease, which limits their usefulness in diagnosing or treating patients; DYNA can accurately perform this task.

To develop DYNA, the investigators applied a type of AI known as a Siamese neural network to fine-tune two existing AI models. They used these modified models to predict the probability that specific gene variants are connected to cardiomyopathy (enlargement, stiffening or weakening of the heart muscle) and arrhythmia (irregular heartbeat).

The investigators then compared the DYNA findings to data in an authoritative public database, known as ClinVar, that archives reports of genetic variations classified for diseases. The data showed that DYNA correctly paired the genetic variants with the given diseases.

“For researchers, DYNA provides a flexible framework to study various genetic diseases,” said Jason Moore, PhD, another contributing author of the study and chair of the Department of Computational Biomedicine at Cedars-Sinai. “Future developments could include using DYNA to offer healthcare professionals advanced tools for tailoring diagnoses and treatments to each individual’s genetic profile.”

The DYNA code is available at GitHub.

Reference: Zhan H, Moore JH, Zhang Z. A disease-specific language model for variant pathogenicity in cardiac and regulatory genomics. Nat Mach Intell. 2025. doi: 10.1038/s42256-025-01016-8

This article has been republished from the following materials. Note: material may have been edited for length and content. For further information, please contact the cited source. Our press release publishing policy can be accessed here.