GLESKit: an MCP-based system for gene-level biomedical investigation and evidence synthesis

GLESKit, an MCP-based gene-centered system, completed all 15 GeneReportBench reports with a mean committee score of 0.904 and answered 35 of 39 BixBench questions.

• Why it matters: Gene-level biomedical work depends on stitching together heterogeneous data, analyses, and literature into reports, but doing so is hard to reproduce and review without bioinformatics expertise.
• What they did: GLESKit links expression, functional annotations, pathways, interaction networks, genetic associations, drug evidence, and literature in one study workspace using Model Context Protocol, then was tested on 15 curated GeneReportBench tasks and a gene-level subset of BixBench.
• The result: Evidence quality varied across reports, while BixBench performance averaged 0.438; in paired tests, adding skills improved scores by 0.145 across 20 questions and 0.085 across five reports.

AI summary of the abstract · may contain errors · not medical advice · read the paper.