benchmark Biohub biology cell biology cell imaging computer vision fluorescence imaging image-based profiling life sciences machine learning microscopy zarr
DynaCell is an evaluation framework for dynamic 3D virtual staining of live cells. The dataset pairs label-free transmitted-light volumes (phase reconstructions from brightfield z-stacks) with fluorescence ground truth for four organelles (nucleus, cell membrane, endoplasmic reticulum, mitochondria) across three conditions (uninfected, Zika-infected, Dengue-infected).The v1 release contains images of A549 human lung adenocarcinoma cells acquired on the Mantis correlative label-free and light-sheet fluorescence microscope at Biohub (24 OZX stores, 262 FOVs, ~407 GB), split into train and test sets across 4 organelle markers and 3 conditions.A v1.1 release will add the WTC-11 hiPSC component: cells from the Allen Institute hiPSC single-cell image dataset (Viana et al., Nature 2023), reprocessed as paired label-free and confocal fluorescence volumes. The iPSC dataset is redistributed under the Allen Institute Terms of Use (https://www.allencell.org/terms-of-use.html), which requires citation of the original dataset and limits use to noncommercial research.All data are stored as RFC-9 zipped OME-Zarr (.ozx) archives following OME-NGFF v0.5, readable with iohub (https://github.com/czbiohub-sf/iohub). Machine-readable metadata (Croissant JSON-LD with Responsible AI fields, per the NeurIPS Datasets & Benchmarks track) is published at s3://dynacell/v1/metadata/croissant.json.
As needed - v1 (A549) is frozen; future releases will expand the dataset. Registry Entry Added 2026-05-04, Last Modified 2026-06-15.
https://github.com/mehta-lab/VisCy/tree/dynacell-models/applications/dynacell
See all datasets managed by Biohub.
DynaCell was accessed on DATE from https://registry.opendata.aws/dynacell.
arn:aws:s3:::dynacellus-west-2aws s3 ls --no-sign-request s3://dynacell/