astrocytes

💫 OpenAstrocytes: Open data and models for astrocyte dynamics

A Python library for discovering, loading, and processing experimental imaging datasets from astrocyte neuroscience research using cloud-hosted data infrastructure.

—❤️‍🔥 Forecast

Features

• Unified Data Discovery: Access experimental datasets through a single Hive interface backed by cloud-hosted manifests
• Type-Safe Schemas: Strongly-typed dataclasses for different experiment types (bath application, photochemical uncaging)
• Lens Transformations: Composable data pipelines for converting raw frames to typed experiments
• atdata + WebDataset Format: Streaming-friendly, schematized TAR archives for efficient cloud storage and access

To see OpenAstrocytes in action, check out the demo in our release pub.

Installation

# Install the core package
pip install astrocytes

# Or with uv (recommended for development)
uv pip install astrocytes

Requirements: Python 3.12 or 3.13

Quick Start

import astrocytes

# Access the data repository
hive = astrocytes.Hive()

# Load a dataset via shortcuts
dataset = astrocytes.data.bath_application

# Iterate through frames
for frame in dataset.ordered(batch_size=None):
    print(f"Frame at t={frame.t:.1f}s, compound={frame.applied_compound}")
    # frame.image is a numpy array of raw 2P imaging data

Architecture

Three-Tier Data Organization

The library organizes imaging data in three tiers:

┌─────────────────────────────────────────────────┐
│  Tier 1: Generic (toile.Frame)                 │
│  Raw imaging data with minimal structure       │
└─────────────────┬───────────────────────────────┘
                  │ Lens Transformation
┌─────────────────▼───────────────────────────────┐
│  Tier 2: Typed Experiments                     │
│  BathApplicationFrame, UncagingFrame, etc.     │
│  Domain-specific metadata extracted            │
└─────────────────┬───────────────────────────────┘
                  │
┌─────────────────▼───────────────────────────────┐
│  Tier 3: Derived Results (Pre-computed)        │
│  EmbeddingResult, EmbeddingPCResult            │
│  Vision transformer outputs, PCA projections   │
└─────────────────────────────────────────────────┘

The Hive Pattern

The Hive class serves as the main entry point, fetching a YAML manifest from the cloud and organizing datasets hierarchically:

hive = astrocytes.Hive()  # Fetches default manifest from data.forecastbio.cloud

# Navigate the hierarchy
generic_frames = hive.index.generic.bath_application.dataset
embeddings = hive.index.embeddings.bath_application.dataset  # Pre-computed embeddings
pca_reduced = hive.index.patch_pcs.bath_application.dataset  # Pre-computed PCA projections

Usage Examples

Working with Typed Experiments

Convert generic frames to experiment-specific types using lens transformations:

import astrocytes
from astrocytes.schema import BathApplicationFrame

# Load generic frames
generic_dataset = astrocytes.data.bath_application

# Apply lens transformation to get typed frames
typed_dataset = generic_dataset.as_type(BathApplicationFrame)

# Now iterate with full type information
for frame in typed_dataset.ordered(batch_size=None):
    print(f"Compound: {frame.applied_compound}")
    print(f"Time: {frame.t:.2f}s (intervention at {frame.t_intervention}s)")
    print(f"Mouse: {frame.mouse_id}, Slice: {frame.slice_id}")
    print(f"Image shape: {frame.image.shape}")
    print(f"Pixel scale: {frame.scale_x}μm × {frame.scale_y}μm")

Working with Pre-computed Embeddings

The data repository includes pre-computed vision transformer embeddings and PCA projections. You can access these directly or apply custom transformations:

from astrocytes import data

# Access pre-computed embeddings
embeddings = data.bath_application_embeddings
for result in embeddings.ordered(batch_size=None):
    print(f"CLS embedding shape: {result.cls_embedding.shape}")
    print(f"Patch embeddings shape: {result.patches.shape}")  # (h, w, embedding_dim)
    break

# Access pre-computed PCA projections
pca_results = data.bath_application_patch_pcs
for result in pca_results.ordered(batch_size=None):
    print(f"Patch PCs shape: {result.patch_pcs.shape}")  # (h, w, n_components)
    break

Experiment Types

Bath Application

Experiments where compounds are applied to the bath solution:

from astrocytes.schema import BathApplicationFrame, BathApplicationCompound

# Compounds: 'baclofen', 'tacpd', 'unknown'
for frame in typed_dataset.ordered(batch_size=None):
    if frame.applied_compound == 'baclofen':
        # Analyze GABA_B receptor activation
        pass
    # ...

Photochemical Uncaging

Experiments using two-photon photo-uncaging to release caged neurotransmitters:

from astrocytes.schema import UncagingFrame

dataset = astrocytes.data.uncaging
typed = dataset.map(UncagingFrame.from_generic)

# Compounds: 'gaba', 'glu', 'laser_only', 'unknown'
for frame in typed.ordered(batch_size=None):
    if frame.uncaged_compound == 'glu':
        # Analyze glutamate uncaging response
        pass
    # ...

Dataset Shortcuts

For convenience, common dataset combinations are available directly:

import astrocytes

# Generic datasets (toile.Frame)
astrocytes.data.bath_application
astrocytes.data.uncaging

# Derived datasets (processed)
astrocytes.data.bath_application_embeddings   # EmbeddingResult
astrocytes.data.bath_application_patch_pcs    # EmbeddingPCResult

Development Setup

# Clone the repository
git clone https://github.com/forecast-bio/open-astrocytes.git
cd open-astrocytes

# Install with development dependencies using uv
uv sync --locked --all-extras --dev

# Run tests
uv run pytest

# Run tests with coverage
uv run pytest --cov=astrocytes --cov-report=html

Project Structure

open-astrocytes/
├── src/astrocytes/
│   ├── __init__.py              # Main package entry point
│   ├── schema.py                # Public schema API
│   └── _datasets/               # Dataset management
│       ├── __init__.py          # Hive and DatasetIndex
│       ├── _common.py           # Base classes
│       ├── _bath_application.py # Bath application schema
│       ├── _uncaging.py         # Uncaging schema
│       ├── _embeddings.py       # Embedding schemas
│       └── _future.py           # Future expansions
├── tests/                       # Test suite
├── pyproject.toml               # Project metadata
└── README.md                    # This file

Key Dependencies

• atdata: Core dataset abstraction and lens transformations
• toile: Generic imaging frame schema
• matplotlib: Plotting and visualization
• scikit-image: Image processing utilities
• scipy: Scientific computing tools

Data Repository

The default data repository is hosted at:

https://data.forecastbio.cloud/open-astrocytes/

The manifest is automatically fetched when you create a Hive() instance. You can specify a custom repository location to use a separate, cloned instance:

hive = astrocytes.Hive(root='https://my-custom-repo.com/astrocytes')

Contributing

Contributions are welcome! To add a new experiment type:

1. Create a new schema module in src/astrocytes/_datasets/_your_experiment.py
2. Define a typed frame class inheriting from ExperimentFrame
3. Implement the from_generic() lens transformation
4. Add the dataset to DatasetIndex in _datasets/__init__.py
5. Export types in schema.py
6. Add tests in tests/test_datasets.py

See CLAUDE.md for detailed development guidelines.

Citation

If you use this library in your research, and please cite:

@article{levesque2025openastrocytes,
  author = {Maxine Levesque and Kira Poskanzer},
  title = {OpenAstrocytes},
  journal = {Forecast Research},
  year = {2025},
  note = {https://forecast.bio/research/open-astrocytes/},
}

License

This project is licensed under the Mozilla Public License 2.0 - see the LICENSE.md file for details.

Acknowledgments

Developed by the Open Science team Forecast.

Docs and README largely by Claude. If they hallucinated, let us know in the Issues!

Support for the production of OpenAstrocytes at Forecast was generously provided by the Special Initiatives division of the Astera Institute.