AI Tool Speeds Up Genetic Diagnosis by Translating Lab Data Into Plain Language
Researchers at Texas Children's and Baylor College of Medicine have developed a computational tool that lets doctors ask questions about genetic variants in everyday language and receive evidence-based answers in seconds, rather than hours.
The tool, called MARRVEL-MCP, combines large language models like ChatGPT and Gemini with structured biological databases to interpret whether genetic changes cause disease or are harmless. The study appeared in the American Journal of Human Genetics on May 22, 2026.
The Problem With Current Methods
Diagnosing rare genetic diseases means determining which DNA changes actually cause disease and which ones don't. Doctors must manually search through multiple databases with different formats, each containing different types of information-how common a variant is, whether it's been linked to disease before, lab results, and relevant research papers.
"Even for experts, this can take hours for a single case," said Dr. Zhandong Liu, chief of computational sciences at Texas Children's.
The team's earlier tool, MARRVEL, already condensed this work from days to minutes by aggregating data sources into one platform. But it required users to input precisely formatted queries and interpret complex outputs-a barrier for non-specialists.
How MARRVEL-MCP Works Differently
The new MCP-based system removes that technical barrier. Instead of learning database formats, users ask questions in plain language: "Is this BRCA1 mutation linked to cancer?"
The system automatically identifies key information, converts it into database formats, queries multiple sources in the correct order, and synthesizes results into a clear answer. The process takes seconds.
Dr. Hyun-Hwan Jeong, an investigator at Texas Children's Duncan Neurological Research Institute, noted that smaller language models perform far better with the right tools. One smaller model improved from 41% accuracy to 94% accuracy when paired with MARRVEL-MCP-and can run locally without expensive cloud infrastructure.
Availability and Next Steps
The researchers released MARRVEL-MCP as an open resource. Users can test it at https://chat.marrvel.org without installing software locally.
The team plans to add these agentic AI features-allowing the system to take independent actions rather than just respond to prompts-directly into the main MARRVEL platform. This would let users move from asking questions to getting structured genetic analysis automatically.
Your membership also unlocks:
