DisPhaseDB 2.0: Improved interpretation of disease‐associated variants in liquid–liquid phase separation proteins with agent‐accessible querying

Membraneless organelles formed through liquid-liquid phase separation (LLPS) are fundamental to cellular organization and are involved in multiple processes, including responses to stimuli and stress. The study of disease-associated variants in LLPS proteins remains vital for understanding protein dysfunction in human diseases. However, maintaining specialized, integrative resources is often hindered when database updates rely on human intervention, typically resulting in long intervals between updates. Here, we introduce a major update to DisPhaseDB (https://disphasedb.leloir.org.ar/), a comprehensive resource for disease-associated variants in LLPS proteins, integrated with an open-source Snakemake workflow organizing systematic data acquisition and parsing into traceable steps. Crucially, the automated system continuously fetches data from source databases, keeping DisPhaseDB up-to-date without the delays of manual maintenance. The updated release expands the database with additional proteins, increases disease annotation coverage by 174%, and adds clinical significance and allele frequency annotations to enhance variant interpretation. To improve accessibility, we also introduce a Model Context Protocol (MCP) server that establishes a standardized interoperability layer, enabling AI agents and large language models to directly query database records through structured operations. This architecture grounds generative workflows in a trusted source, replacing unconstrained web retrieval and reducing unsupported content. In a comparative benchmark, data retrieval through the MCP server achieved a mean F1 score of 0.99, compared to 0.30 for unguided generative retrieval. Together, these developments position DisPhaseDB2.0 as a maintainable resource for LLPS-related variants, optimizing reproducible data access for both human researchers and emerging agentic AI workflows.