# DynaCell v1

An evaluation framework for dynamic 3D virtual staining of live cells.

## Directory structure

```
dynacell_v1/
|-- data/
|   |-- biohub-a549/       A549 Mantis data (CC-BY-4.0)
|   |   |-- train/         Training split (.ozx OME-Zarr stores)
|   |   `-- test/          Test split (.ozx OME-Zarr stores)
|   `-- aics-hipsc/        iPSC WTC-11 data (Allen Institute Terms of Use)
|-- demo/                  Reviewer-accessible sample (< 4 GB)
|-- metadata/              Croissant RAI metadata and QC reports
|-- models/                Trained checkpoints and configs
`-- supplementary/         Supplementary movies
```

## Files

The full release contains 30 OME-Zarr archives (`.ozx`), ~677 GB total. Each
file is downloadable directly via HTTPS as
`https://dynacell.s3.us-west-2.amazonaws.com/v1/<path>`; bucket browsing
requires the AWS CLI until the AWS Open Data registry PR
(<https://github.com/awslabs/open-data-registry/pull/3130>) merges:

```bash
aws s3 ls --no-sign-request s3://dynacell/v1/
aws s3 cp --no-sign-request --recursive s3://dynacell/v1/data/biohub-a549/test/ ./
```

### data/biohub-a549/ — 387 GB (CC-BY-4.0)

4 organelles × 3 conditions × 2 splits = 24 archives.

```
train/CAAX_mock.ozx       30G    test/CAAX_mock.ozx       13G
train/CAAX_DENV.ozx       15G    test/CAAX_DENV.ozx       13G
train/CAAX_ZIKV.ozx       30G    test/CAAX_ZIKV.ozx       13G
train/H2B_mock.ozx        30G    test/H2B_mock.ozx        13G
train/H2B_DENV.ozx        15G    test/H2B_DENV.ozx        13G
train/H2B_ZIKV.ozx        29G    test/H2B_ZIKV.ozx        13G
train/SEC61B_mock.ozx     23G    test/SEC61B_mock.ozx     11G
train/SEC61B_DENV.ozx    3.0G    test/SEC61B_DENV.ozx    9.7G
train/SEC61B_ZIKV.ozx     23G    test/SEC61B_ZIKV.ozx     11G
train/TOMM20_mock.ozx     21G    test/TOMM20_mock.ozx     11G
train/TOMM20_DENV.ozx    5.3G    test/TOMM20_DENV.ozx     11G
train/TOMM20_ZIKV.ozx     21G    test/TOMM20_ZIKV.ozx     11G
```

### data/aics-hipsc/ — 290 GB (Allen Institute Terms of Use)

3 archives × 2 splits = 6 archives. `cell.ozx` carries paired Membrane and
Nucleus channels alongside a single brightfield input.

```
train/cell.ozx            86G    test/cell.ozx           7.0G
train/SEC61B.ozx          84G    test/SEC61B.ozx         8.1G
train/TOMM20.ozx          97G    test/TOMM20.ozx         7.9G
```

### demo/ — 745 MB

`demo/dynacell_a549_demo.zip` — reviewer-accessible bundle (OZX + single-cell
embeddings + scripts + README). See `demo/README.md`.

## Licenses

- **Biohub A549 data**: Creative Commons Attribution 4.0 International (CC-BY-4.0).
- **Allen Institute iPSC data**: Allen Institute Terms of Use (noncommercial research). See `data/aics-hipsc/README.md`.
- **Benchmark code**: BSD-3-Clause. Source: <https://github.com/mehta-lab/VisCy>.

## Hosting

Data are hosted on AWS Open Data at `s3://dynacell/v1/`.

## Contact

Shalin B. Mehta (shalin.mehta@czbiohub.org), Alexandr A. Kalinin (alex.kalinin@czbiohub.org).

Code: https://github.com/mehta-lab/VisCy/pull/404, 
Issues: <https://github.com/mehta-lab/VisCy/issues>