In the span of seconds, a doctor in Houston can now ask an AI a question in plain English—"Is this BRCA1 mutation linked to cancer?"—and receive a clear, evidence-based answer that once would have taken hours of database hunting and expert interpretation. This shift marks a meaningful change in how genetic diagnosis works, powered by a new computational tool called MARRVEL-MCP developed by researchers at Baylor College of Medicine and Texas Children's Hospital.
Rare genetic diseases are often caused by tiny changes in a person's DNA, but not every genetic change linked to a condition actually causes it—some variants are innocent bystanders. For clinicians and researchers trying to separate the harmful mutations from the harmless noise, the current process is exhausting. Doctors must gather information from many different biological databases, each with its own format and rules, then piece together the evidence like a complex puzzle. Even for experts, this can consume hours for a single patient case. The stakes are high: misidentifying a variant means families may never get a diagnosis, while false positives can lead them down the wrong treatment paths.
MARRVEL-MCP builds on earlier work. Baylor and Texas Children's previously developed MARRVEL (Model organism Aggregated Resources for Rare Variant ExpLoration), a tool that let researchers search through large genetic and biological databases in minutes instead of days. The uptake has been significant—MARRVEL recorded more than 43,000 users worldwide in 2025 alone, bringing together genomic, functional, and model-organism databases into one platform. But MARRVEL demanded precisely formatted technical inputs and produced complex outputs that required substantial expertise to interpret. It worked well for specialists; it wasn't built for the broader medical community.
MARRVEL-MCP changes that equation by integrating artificial intelligence—specifically large language models like ChatGPT and Gemini—with those curated biological databases. Instead of learning technical formats and navigating multiple sources manually, users simply ask questions in everyday language. The AI automatically identifies key pieces of information (gene names, mutations), converts them into the formats databases require, queries multiple sources in the correct order, and synthesizes the results into a single, clear answer. Within seconds, not hours.
The system covers disease associations, genetic variation, gene expression, and scientific literature, enabling these smaller AI models to autonomously compose and execute multi-step analytical workflows from plain-language queries. This matters because it democratizes access to genetic diagnosis tools. A clinician in a rural hospital or a non-expert researcher can now tap into the same depth of biological knowledge that once required a specialized genomicist working for hours at a terminal.
Dr. Hyun-Hwan Jeong, assistant professor of pediatrics–neurology at Baylor and an investigator at the Duncan Neurological Research Institute at Texas Children's, notes that the real excitement lies not in deploying the largest, most sophisticated AI models available, but in giving smaller models access to the right tools and structured context. That approach is both more sustainable and more democratizing than building ever-larger systems. A study describing MARRVEL-MCP was published in the American Journal of Human Genetics, opening the door for other institutions to adopt and build on the platform as the work of genetic diagnosis continues to accelerate.