Translating Genetic Complexity
Researchers at the Texas Children’s Duncan Neurological Research Institute and Baylor College of Medicine have developed a computational tool named MARRVEL-MCP. This software bridges the gap between massive biological databases and clinical diagnosis by allowing users to query genetic data using plain language. The study detailing this advancement appeared in the American Journal of Human Genetics on May 22, 2026.
Rare genetic diseases often stem from small DNA alterations. Determining which specific variation causes a condition remains a difficult task for medical professionals. Doctors currently spend hours manually gathering evidence from multiple databases, each governed by different technical requirements. This process creates a bottleneck for clinical diagnosis.
The Technical Shift
MARRVEL-MCP builds upon the team’s existing platform, MARRVEL, which launched previously to aggregate model organism data and genomic information. While the original system helped over 43,000 users in 2025 alone, it required specialized formatting and manual synthesis of output. The new protocol removes these friction points.
By integrating large language models with curated databases, the system automates complex workflows. A user simply asks a question like whether a specific BRCA1 mutation links to cancer. The system handles the conversion of that query into a database-ready format, retrieves the evidence, and generates a clear answer in seconds. It covers areas such as gene expression, disease association, and existing scientific literature.
Future Implications for Rare Disease Research
Dr. Hyun-Hwan Jeong notes that the team proved smaller artificial intelligence models can achieve high accuracy when provided with the correct context. When testing a locally installed model called gpt-oss-20b, the researchers observed an accuracy jump from 41% to 94% after applying MARRVEL-MCP. This result indicates that labs do not always need massive, expensive models to achieve significant results in biomedical research.
The team has released the tool as an open resource. A public interface is available for researchers to test the system without local installation. The developers plan to add agentic features in future updates. These additions would allow the software to perform independent, multi-step tasks, moving beyond simple text responses to provide full, automated genetic analysis.
Support for this project came from several institutions, including the Cancer Prevention and Research Institute of Texas and the National Institutes of Health. This work represents a shift toward more accessible, cost-effective digital tools in pediatric and neurological medicine.