ChatAnalysis: Galaxy Data Analysis Agent

Introduction

ChatAnalysis is an interactive Galaxy tool that lets you perform data analysis via natural language. Built on Large Language Models (LLMs) and the DSPy prompt framework, ChatAnalysis interprets plain-English requests (e.g. “Generate a heatmap of columns A vs B”) and returns:

• Data processing (filtering, aggregations, summary statistics)
• Visualizations (plots, charts)
• Code snippets (Python / Pandas / AutoGluon)
• Explanations (how the analysis was done)

All results appear in a split-pane chat interface inside Galaxy—no coding required!

---

v0.3 Highlights (DSPy Agent)

• DSPy + Few-Shot Learning
  Uses curated Q&A examples to improve prompt accuracy.

• Split-Pane UI

  • Left: Conversation
  • Right: Details & Preview (code, tables, plots)
    Click past queries to revisit full outputs.

• Faster, Multi-Step Analyses
  Optimized logic for chaining data cleaning, visualization, modeling in one session.

• Advanced ML Support
  Integrates AutoGluon, scikit-learn, and more for model training, evaluation, and explanation.

• Bookmarking & History
  Easily revisit previous queries without rerunning; suggested follow-up shortcuts.

---

Use on Galaxy Main (usegalaxy.org)

1. Upload Data
   Log in to usegalaxy.org, upload your CSV/TSV/Excel file into your history.

2. Set Your LLM API Key

   • Obtain an OpenAI API key (or other supported backend).
   • In Galaxy: User → Preferences → OpenAI API Key → paste your key.

3. Launch ChatAnalysis

   • Find ChatAnalysis under Interactive Tools.
   • Select your dataset (and model if prompted) → Launch.

4. Chat & Analyze

   • Type queries like “Summarize this dataset.”
   • View text, code, tables, and plots in the split-pane UI.

5. Follow-Ups & Refinement

   • Context is retained: ask “Filter to 2022 and rerun.”
   • Click suggested follow-up buttons for one-click analyses.

6. Stop & Save

   • Click Stop to end the session.
   • Outputs are saved back to your Galaxy history.

---

Install on Your Own Galaxy

1. ToolShed Installation (Admin)

• Repo: goeckslab/chatanalysis
• Install via ToolShed; enable Interactive Tool support and Docker.

2. Local Development

git clone https://github.com/goeckslab/ChatAnalysis.git
cd ChatAnalysis

# 1. Create & activate a virtual environment
python3.11 -m venv .env
source .env/bin/activate

# 2. Install dependencies
pip install --upgrade pip
pip install -r dspy_agent/requirements_nicegui_dspy.txt

# 3. Provide your OpenAI API key
# Option A: Key file (default)
echo YOUR_OPENAI_KEY > dspy_agent/user_config_openai.key

# Option B: CLI argument
# python dspy_agent/chat_dspy.py --openai_key_file /full/path/to/user_config_openai.key

# Option C: provide on the app UI.

# 4. Run the DSPy agent
python dspy_agent/chat_dspy.py

# 5. Open in browser:
# http://localhost:9090

or

docker build -t chatanalysis:dspy -f dspy_agent/Dockerfile .
docker run -p 9090:9090 chatanalysis:dspy
# then visit http://localhost:9090

Note: Internet access is required for the LLM API.

---

Legacy Agents (Reference Only)

These prior versions are now deprecated—moved to legacy_agents/:

• v0.1 – pandasai_agent/
• v0.2 – smolagents_agent/

---

Repository Structure

/
├── .github/workflows/        # CI/CD pipelines
├── dspy_agent/               # **v0.3 DSPy agent** (primary code)
├── legacy_agents/            # pandasai_agent/ & smolagents_agent/
├── tools/                    # Galaxy tool XML & support files
├── seq_diagram.md            # Sequence diagram
├── LICENSE
├── README.md
├── .gitignore
└── .dockerignore

---

Feel free to open issues or pull requests for questions or improvements!