AI Tool Accelerates Time to Rare Diagnosis

Rare Daily Staff

Researchers at Baylor College of Medicine and Texas Children’s Hospital have developed an AI-powered tool designed to speed up and simplify the process of diagnosing rare genetic diseases, according to a study published in the American Journal of Human Genetics.

The tool, called MARRVEL-MCP, uses large language models alongside curated biological databases to help interpret genetic variants through plain-language queries. The approach is designed to reduce the time and level of expertise required to determine whether a specific genetic mutation is clinically relevant.

“Identifying whether a particular genetic change is harmful or an innocent bystander is crucial for diagnosing these conditions, but the process requires sifting through large amounts of data,” said Hyun-Hwan Jeong, assistant professor of pediatrics–neurology at Baylor and co-corresponding author of the study.

Diagnosing rare genetic diseases typically involves aggregating evidence from multiple databases, each with different formats and rules. Researchers must evaluate factors such as how common a variant is, whether it has been previously linked to disease, and what functional data from laboratory or model organism studies suggest. Even for experienced clinicians and scientists, this process can take hours per case.

MARRVEL-MCP builds on the team’s earlier platform, MARRVEL, which already integrates diverse genomic and biological datasets into a single interface. While MARRVEL has attracted tens of thousands of users globally, it still requires technical expertise to input queries correctly and interpret results.

The new system removes many of those barriers. Users can ask questions in everyday language, and the tool automatically structures the query, searches multiple databases, and synthesizes the findings into a clear, evidence-based response.

Zhandong Liu, associate professor at Baylor and chief of computational sciences at Texas Children’s, said the goal is to make advanced genomic analysis more accessible to non-specialists. The platform is available through a public interface, allowing users to test the system without installing software locally.

The team has released MARRVEL-MCP as an open resource and plans to expand its capabilities by incorporating more autonomous features into the broader MARRVEL platform. These enhancements could allow the system to move beyond answering questions to initiating structured analyses based on user input.