LA-MARRVEL: A Knowledge-Grounded, Language-Aware LLM Framework for Clinically Robust Rare Disease Gene Prioritization

Area of Science:

• Genomics
• Artificial Intelligence
• Clinical Diagnostics

Background:

• Rare disease diagnosis is complex, requiring extensive gene-phenotype matching.
• Current clinical interpretation pipelines for rare diseases are time-intensive.
• Accurate gene prioritization is crucial for efficient rare disease diagnosis.

Purpose of the Study:

• To introduce LA-MARRVEL, a novel language-aware Large Language Model (LLM) framework for robust and practical rare disease diagnosis.
• To enhance gene prioritization accuracy and clinical interpretation workflows.
• To provide auditable, clinically relevant reasoning for candidate gene prioritization.

Main Methods:

• Developed LA-MARRVEL, a knowledge-grounded, language-aware LLM framework.
• Utilized structured, phenotype-rich prompt construction encoding patient and disease phenotypes.
• Evaluated performance across three real-world cohorts and compared against established gene prioritization approaches.

Main Results:

• LA-MARRVEL achieved a 12-15 percentage-point absolute improvement in Recall@1 over existing methods.
• Structured prompt construction was identified as a key factor for accuracy gains.
• Consistent improvement in gene-ranking performance was observed, including in challenging cases.

Conclusions:

• LA-MARRVEL significantly enhances gene prioritization in rare disease diagnosis.
• Phenotype-rich prompt engineering is critical for preserving clinical context.
• The framework offers a viable method to augment existing diagnostic pipelines with auditable explanations.