CellVoyager

https://github.com/zou-group/cellvoyager
Admin
CellVoyager is an AI-powered agent and interactive interface designed for automated hypothesis...
Tokens:7,260
Snippets:72
Trust Score:9
License:MIT
Update:1 month ago
Show doc for...
Tokens:
Raw
### Launch GUI

Source: https://github.com/zou-group/cellvoyager/blob/main/README.md

Start the browser-based Streamlit interface.

```bash
streamlit run gui/app.py
```

--------------------------------

### Install CellVoyager Environment

Source: https://github.com/zou-group/cellvoyager/blob/main/README.md

Clone the repository and set up the required Conda environment.

```bash
git clone https://github.com/zou-group/CellVoyager.git
cd CellVoyager
conda env create -f environment.yml
conda activate CellVoyager
```

--------------------------------

### View Help Options

Source: https://github.com/zou-group/cellvoyager/blob/main/README.md

Display the full list of available command-line arguments.

```bash
python run_cellvoyager.py --help
```

--------------------------------

### Configure API Keys

Source: https://github.com/zou-group/cellvoyager/blob/main/README.md

Create a .env file in the project root to store your OpenAI and Anthropic API keys.

```text
OPENAI_API_KEY=sk-xxxxxxxxxxxxx
ANTHROPIC_API_KEY=sk-ant-xxxxxxxxxxxxx
```

--------------------------------

### Run agent on CellBench

Source: https://github.com/zou-group/cellvoyager/blob/main/README.md

Executes the agent three times on CellBench using either gpt-4o or o3-mini.

```bash
cd CellBench
python run_agent.py {gpt-4o|o3-mini}
```

--------------------------------

### Run base LLMs on CellBench

Source: https://github.com/zou-group/cellvoyager/blob/main/README.md

Executes base LLMs (gpt-4o, o3-mini) three times on the CellBench dataset.

```bash
cd CellBench
python run_base_llm.py
python run_llm_judge.py
```

--------------------------------

### System Prompt Template Definition

Source: https://github.com/zou-group/cellvoyager/blob/main/cellvoyager/prompts/coding_system_prompt.txt

Defines the system prompt string for the CellVoyager agent, enforcing a maximum step count and a specific JSON output schema.

```python
system_prompt = f"""
        You are a creative and skilled expert in single-cell transcriptomics computational analysis.

        CRITICAL: The analysis plan MUST NOT exceed {max_iterations} steps. You are strictly limited to exactly {max_iterations} steps or fewer. Do not suggest more.

        Output your response in the following JSON format (do not number the analysis steps, just list them):
        {{
            "hypothesis": "...",
            "analysis_plan": ["First step", "Second step", ...],
            "first_step_code": "...",
            "code_description": "1-2 sentences describing the goal of the code, what it is doing, and why",
            "summary": "1-2 sentence summary of the analysis. Just return what the analysis is doing overall. Don't note changes."
        }}
        """
```

--------------------------------

### Run Terminal Analysis

Source: https://github.com/zou-group/cellvoyager/blob/main/README.md

Execute an analysis directly from the terminal by providing the dataset path, paper summary, and analysis name.

```bash
python run_cellvoyager.py --h5ad-path PATH_TO_H5AD_DATASET \
                          --paper-path PATH_TO_PAPER_SUMMARY \
                          --analysis-name RUN_NAME
```

--------------------------------

### Run default neighborhood graph computation

Source: https://github.com/zou-group/cellvoyager/blob/main/cellvoyager/prompts/DeepResearch_Analyses.txt

Executes the default neighborhood graph construction, which is a prerequisite for UMAP visualization.

```python
sc.pp.neighbors(adata)
sc.tl.umap(adata)
```

--------------------------------

### Compute k-nearest-neighbor graph with sc.pp.neighbors

Source: https://github.com/zou-group/cellvoyager/blob/main/cellvoyager/prompts/DeepResearch_Analyses.txt

Constructs a kNN graph on the low-dimensional representation, populating connectivities and distances in the AnnData object.

```python
sc.pp.neighbors(adata, n_neighbors=15, n_pcs=30)
```

--------------------------------

### Download COVID-19 dataset

Source: https://github.com/zou-group/cellvoyager/blob/main/README.md

Downloads the .h5ad dataset file from the specified S3 bucket using curl.

```bash
curl -o example/covid19.h5ad "https://hosted-matrices-prod.s3-us-west-2.amazonaws.com/Single_cell_atlas_of_peripheral_immune_response_to_SARS_CoV_2_infection-25/Single_cell_atlas_of_peripheral_immune_response_to_SARS_CoV_2_infection.h5ad"
```

--------------------------------

### Extract latent representation from scVI model

Source: https://github.com/zou-group/cellvoyager/blob/main/cellvoyager/prompts/DeepResearch_Analyses.txt

Extracts the probabilistic latent space representation from a trained scVI model and stores it in the AnnData object.

```python
latent = model.get_latent_representation()
adata.obsm['X_scVI'] = latent
```

--------------------------------

### Compute UMAP embedding with Scanpy

Source: https://github.com/zou-group/cellvoyager/blob/main/cellvoyager/prompts/DeepResearch_Analyses.txt

Performs UMAP dimensionality reduction on an AnnData object, typically after computing the neighborhood graph.

```python
sc.tl.umap(adata)
```
Tokens:
Raw