OpenAstrocytes

📰 Source for reproducing the pub frontend

This repository contains the code and analysis for the OpenAstrocytes pub, demonstrating how off-the-shelf vision transformer embeddings (DINOv3) capture rich structure in astrocyte calcium dynamics.

The analysis demonstrates that DINOv3 patch embeddings applied to astrocyte calcium imaging data can:

• Reflect astrocyte network anatomy
• Capture evoked dynamics in response to compound applications
• Generalize across distinct experimental sessions and subjects
• Enable accurate decoding of applied compounds from single patches

—❤️‍🔥 Forecast

Overview

This project combines:

• A Python package (oa_pub) with analysis utilities for working with astrocyte calcium imaging and embeddings
• A Quarto document that generates the full pub static site with embedded analysis, figures, and narrative

Repository Structure

pub-open-astrocytes/
├── src/oa_pub/              # Python analysis package
│   ├── analysis/            # Data loading and statistical utilities
│   │   ├── _common.py      # Helper functions (e.g., is_between)
│   │   ├── _constants.py   # Constants (patch grid dimensions, compound aliases)
│   │   ├── _extraction.py  # Data loading (get_movie, get_movie_traces)
│   │   └── _stats.py       # Statistics (baseline_normalize, bootstrap)
│   └── figures/             # Plotting utilities
│       └── __init__.py     # Visualization functions
├── quarto/                  # Publication source
│   ├── index.qmd           # Main Quarto document with analysis
│   ├── _quarto.yml         # Quarto configuration
│   ├── _brand.yml          # Forecast branding
│   ├── img/                # Images and assets
│   └── csl/                # Citation styles
├── _public/                 # Generated publication output
├── pyproject.toml          # Python package configuration
└── README.md               # This file

Prerequisites

• Python 3.12 or later
• uv package manager (installation instructions)
• Quarto (installation instructions)

Installation

1. Clone the repository

git clone https://github.com/forecast-bio/pub-open-astrocytes.git
cd pub-open-astrocytes

2. Set up Python environment

# Install dependencies with uv
uv sync

# This creates a virtual environment in .venv/ and installs all dependencies

3. Verify installation

# Test that the package is importable
uv run python -c "import oa_pub; print('Success!')"

Reproducing the pub

Generate the full pub

cd quarto
uv run quarto render

This will:

• Execute all Python code blocks in index.qmd
• Load data from the OpenAstrocytes dataset
• Generate all figures and analyses
• Output the complete pub to _public/index.html

Note: The first render will take significant time as it streams data from the cloud and computes embeddings. Subsequent renders use Quarto's caching system for faster execution.

Preview with live reload

cd quarto
uv run quarto preview

This opens a browser with the pub and automatically reloads when you make changes to index.qmd.

Working with the Python Package

Using the analysis utilities

import oa_pub.analysis as analysis

# Load a recording by UUID
movie_uuid = "your-movie-uuid-here"
frames = analysis.get_movie(movie_uuid, verbose=True)

# Get patch embedding traces (14×14 grid)
traces = analysis.get_movie_traces(movie_uuid, verbose=True)

# Extract time information
times, dt = analysis.movie_times(frames)

# Apply baseline normalization
normalized, mean, std = analysis.baseline_normalize(
    times, signal, window=(-10, 0)
)

# Bootstrap statistics
boot_results = analysis.boot_stat(
    data,
    lambda x: np.mean(x, axis=0),
    n=200
)

Using the plotting utilities

from oa_pub.figures import plot_micrograph, plot_trace_grid
from forecast_style import HouseStyle

# Plot a micrograph with scale bar
with HouseStyle() as s:
    plot_micrograph(s, image, scale_bar=100.)

# Plot a grid of traces
with HouseStyle() as s:
    fig, axs = plt.subplots(14, 14, figsize=(12, 12))
    plot_trace_grid(axs, times, traces, color='b')

Data Access

This project uses the OpenAstrocytes dataset, which is publicly available and streamed on-demand:

# Install the astrocytes package
pip install astrocytes

# Or with uv
uv add astrocytes

The oa_pub.analysis module handles data streaming automatically when you call functions like get_movie() and get_movie_traces().

Use verbose parameters for progress output during data loading.

Citation

If you use this code or the OpenAstrocytes dataset in your research, please cite:

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

Links

• OpenAstrocytes data hive: Available via the astrocytes package
• Forecast: https://forecast.bio
• OpenAstrocytes pub: View the rendered pub

License

Copyright © 2025 Forecast Bio, Inc.

The narrative contents of this article are licensed under the Creative Commons CC-BY 4.0 license.

Code contributions in this repo are licensed under the Mozilla Public License 2.0; see the LICENSE.md file for details.

Acknowledgements

Developed by the Open Science team at Forecast.

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

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