New AI Tool Makes Genetic Diagnosis More Accessible for Non-Experts
Researchers at Baylor College of Medicine and Texas Children's Hospital have developed MARRVEL-MCP, a computational tool that uses large language models to interpret genetic variants using plain language.
The tool is designed to answer simple questions like "Is this BRCA1 mutation linked to cancer?" by automatically searching multiple databases and producing evidence-based answers.
How It Works
MARRVEL-MCP builds on the existing MARRVEL platform, which already aggregates genomic, functional, and model-organism databases. The key innovation is its ability to understand natural language queries and autonomously execute multi-step analytical workflows.
Users can input plain-language questions, and the tool automatically identifies relevant information, queries multiple databases, and produces evidence-based answers — dramatically reducing the time and expertise required to analyze variant data.
Performance Breakthrough
The research team reports a significant accuracy improvement when integrating AI with MARRVEL-MCP. A smaller AI model (gpt-oss-20b) improved its accuracy from 41% to 94% when using the tool.
Background & Development
The original MARRVEL platform already allowed researchers to search multiple biological databases simultaneously, but it required precise input and manual interpretation of complex outputs. MARRVEL-MCP addresses these limitations by incorporating AI agents that can handle the entire analytical workflow.
This makes genetic diagnosis more accessible to non-experts by reducing both the time and specialized knowledge needed to interpret variant data.
Publication & Availability
The study was published in the American Journal of Human Genetics. First author Zachary Everton, along with Jorge Botas, Seon Young Kim, and Lin Yao, contributed to the work.
MARRVEL-MCP is available as an open resource at https://chat.marrvel.org.
The project was supported by the Cancer Prevention and Research Institute of Texas, the Chan Zuckerberg Initiative, the National Institutes of Health, and other funding sources.