DeepBiology Lab agent extensions

DeepBiology Lab helps researchers answer fundamental questions about gene regulation using natural-language prompts and LLM agents. These questions can range from "How is this gene regulated?" and "What is the effect of this cancer-associated mutation?" to "Can we devise approaches to reactivate this important tumor suppressor gene in cancer patients?"

This repository provides a Python SDK and agent extensions for Qwen, Gemini, and Codex. By default, these agents connect to a centralized Model Context Protocol (MCP) server that exposes 13 skills powered by the DeepBiology Python SDK. The service is accessed over HTTPS and is deployed on an Alibaba Cloud VM.

DeepBiology Lab assays

DeepBiology Lab currently provides four gene-regulation analysis workflows:

Workflow	Question
Q1 — Regulation	How does predicted transcription change across genomic coordinates for a gene and cell line?
Q2 — Enhancer importance	Which nucleotides in a regulatory region are most important to the predicted signal?
Q3 — Mutation impact	How does a specified DNA sequence change affect predicted regulation or enhancer function?
Q4 — Enhancer redesign	Can an enhancer sequence be redesigned to optimize its predicted activity?

The extension also includes tools for resolving and normalizing gene names, cell-line names, variants, and cancer mutations before job submission.

Installation

To use the extension, configure the environment variables below and install the extension for your agent. Installing the agents themselves is outside the scope of this guide. The Python SDK is optional for workflows that use the centralized MCP service.

Set environment variables

Set the centralized MCP endpoint and your DeepBiology API key before starting an agent. The MCP server uses the API key to authenticate requests. The URL must include /mcp.

Sign in at deepbiology.ai to create your DeepBiology API key. Contact info@deepbiology.ai if you need help requesting an account or creating a DeepBiology API key. To request the MCP URL, contact jinsongzhang@deepbiology.ai. Replace the example MCP hostname below with the endpoint.

export DEEPBIOLOGY_MCP_URL=https://your-mcp-host.example/mcp
export DEEPBIOLOGY_API_KEY=dbio_your_api_key_here

Install the optional DeepBiology Python SDK

Python 3.9 or newer is required for the SDK and optional local tools.

pip install git+https://github.com/DeepBiology/deepbiology-lab.git

To use the SDK or local tools, save your API key in the local configuration:

deepbiology-lab config --api-key dbio_your_api_key_here

This configuration step is not required for agent workflows that use the centralized MCP service.

Install the DeepBiology extension

Qwen CLI

Run the following command and follow the prompts to enter the MCP URL and API key:

qwen extensions install https://github.com/DeepBiology/deepbiology-lab

Gemini CLI

Run the following command and follow the prompts to enter the MCP URL and API key:

gemini extensions install https://github.com/DeepBiology/deepbiology-lab --auto-update

Codex CLI

codex plugin marketplace add DeepBiology/deepbiology-lab
codex plugin add deepbiology@deepbiology-marketplace
codex mcp add deepbiology-lab \
  --url "$DEEPBIOLOGY_MCP_URL" \
  --bearer-token-env-var DEEPBIOLOGY_API_KEY

Run an example workflow with Qwen

qwen

Identify enhancers for the MYC gene in MCF7 cells.

Skills

Skill	What it does
deepbiology-setup	Configure the SDK, API key, and MCP connection
deepbiology-resolve-gene	Resolve gene names and aliases to HGNC symbols
deepbiology-resolve-cell-line	Resolve model- and assay-specific cell-line channels
deepbiology-list-models	List supported model catalogs
deepbiology-resolve-snps	Find regional variants and annotate dbSNP identifiers
deepbiology-cancer-mutations	Query cancer-mutation annotations
deepbiology-q1-regulation	Run Q1 transcription-regulation analysis
deepbiology-q2-enhancer-importance	Run Q2 enhancer-importance analysis
deepbiology-q3-mutation-impact	Run Q3 mutation-impact analysis
deepbiology-q4-enhancer-redesign	Run Q4 enhancer-redesign analysis
deepbiology-check-status	Check a submitted job
deepbiology-get-result	Retrieve and explain a completed result
deepbiology-download-result	Save result JSON and optional image artifacts locally

Architecture

Qwen, Gemini, and Codex share the same skills and a centralized MCP service. The service is deployed on an Alibaba Cloud VM and authenticates every request with the caller's DeepBiology API key.

flowchart LR
    U[User] <--> Q[Qwen CLI]
    U <--> G[Gemini CLI]
    U <--> C[Codex CLI]
    Q <--> S[Shared agent skills]
    G <--> S
    C <--> S
    S <--> M[Centralized MCP server<br/>Alibaba Cloud VM]
    M <--> D[DeepBiology Lab services]

Loading

Notes

• DEEPBIOLOGY_MCP_URL must use the complete HTTPS endpoint, including /mcp.
• Do not commit API keys. Each user must supply their own key.

License

This project is licensed under the MIT License.