Skip to content
LatestRSS
TechBio Today

The front page of AI-driven biology.

peer reviewed · Bioinformatics

Searching the Druggable Genome using Large Language Models

Lars Schimmelpfennig and colleagues report that a Model Context Protocol server lets language models query current, structured drug-gene interaction data from DGIdb.

Diagram showing an LLM querying CIViC for ibrutinib-resistance evidence and DGIdb for drugs that target the resulting genes.
The paper’s worked example chains CIViC and DGIdb queries to connect resistance evidence with candidate drug interactions.Source: Schimmelpfennig et al.

Author affiliations

  • Washington University School of Medicine
  • Nationwide Children’s Hospital
  • The Ohio State University College of Medicine

In plain English

Instead of expecting a model to remember drug-gene relationships, the server translates a supported tool call into DGIdb API access and returns structured database information.

How the study worked

A plain-language walk through the work behind the result.

  1. Built an MCP server around the DGIdb API.

  2. Tested language-model answers to questions requiring current structured biomedical knowledge.

What they found

  • The abstract reports improved answer ability when the MCP server supplied DGIdb information.
  • The publisher abstract does not state a numerical effect size.

Why it matters

Tool-bound database access can make scientific assistants more current and auditable than relying on model memory alone.

The catch

  • The abstract does not report the evaluation-set size or numerical improvement.
  • Correct tool access does not itself validate downstream interpretation.

Evidence ledger

Sources behind this brief

  1. 01
    Primary source

    Bioinformatics accepted manuscript

    peer reviewed · Accessed August 24, 2026

  2. 02
    Supporting context

    DGIdb MCP server repository

    author reported result · Accessed August 24, 2026