MARRVEL-MCP enables natural language variant interpretation through autonomous workflow construction
Everton, Z. J.; Botas, J.; Kim, S. Y.; Liu, Z.; Jeong, H.-H.
Show abstract
AO_SCPLOWBSTRACTC_SCPLOWRare disease variant interpretation requires navigating multiple genomic databases with strict input formats and synthesizing heterogeneous evidence, creating barriers for non-experts and cognitive burdens even for specialists. MARRVEL exemplifies this challenge by requiring precise queries (e.g., HGVS notation) and returning complex, difficult-to-synthesize outputs. To address this input-output asymmetry, we developed MARRVEL-MCP, a natural language interface enabling large language models (LLMs) to perform end-to-end variant interpretation via structured tool access. This work demonstrates the impact of context engineering--the deliberate design of domain-aware tool environments and information scaffolding--in reshaping the role of model scale in genomics. MARRVEL-MCP equips LLMs with 39 tools spanning gene and variant utilities, pathogenicity databases, phenotype resources, expression atlases, ortholog data, and literature APIs. Without hard-coded pipelines, LLMs autonomously infer workflows, performing named-entity recognition, identifier normalization, and multi-database synthesis from narrative queries. Using 45 expert-curated questions, lightweight models (3B-20B parameters) with MARRVEL-MCP matched or outperformed larger models lacking tool access. A 20B-parameter model (gpt-oss-20b) achieved a 95% pass rate versus 33% without MARRVEL-MCP, approaching state-of-the-art proprietary performance. Although accuracy remains below the 100% required for autonomous diagnosis and token usage is considerable, these results show that well-designed context can compensate for limited model capacity. These findings establish context engineering as a core principle for biomedical AI, enabling efficient, scalable integration of LLMs with curated genomic resources. MARRVEL-MCP is available at https://github.com/hyunhwan-bcm/MARRVEL_MCP/.
Matching journals
The top 8 journals account for 50% of the predicted probability mass.
Similar papers in this journal
Similar papers in this journal
Similar papers in this journal
Similar papers in this journal
- Towards a standard benchmark for phenotype-driven variant and gene prioritisation algorithms: PhEval - Phenotypic inference Evaluation framework 96%
- RTX-KG2: a system for building a semantically standardized knowledge graph for translational biomedicine 95%
- SKiM-GPT: Combining Biomedical Literature-Based Discovery with Large Language Model Hypothesis Evaluation 95%
Similar papers in this journal
"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.