How to Make Imaging Data Ready for Reanalysis

The problem — Data that the original author can re-open is not the same as data a stranger can reuse. "Reanalysis-ready" sets the bar at the latter: a group with no contact with the source lab should be able to find the dataset, open it, understand what every channel and dimension means, and re-run or re-interpret it without emailing anyone. Most published imaging data fails this test — it lives on a private drive, in a proprietary format, with the crucial context (pixel size, perturbation, acquisition settings) only in the first author's memory. When that person leaves, the dataset dies.

What it is / how it works — Three layers make data reanalysis-ready. FAIR (Wilkinson et al.) is the target: Findable, Accessible, Interoperable, Reusable — with an explicit emphasis on machine-actionability, so software, not just humans, can locate and consume the data. Metadata completeness is the substance: REMBI (Sarkans et al.) defines the recommended metadata for biological images — study, sample, acquisition, and analysis fields — that a reuser needs and an author rarely volunteers; an open format like OME-NGFF carries much of it in the array itself. Public deposition is the delivery: archives such as the BioImage Archive give the dataset a stable identity and home, so "accessible" is a URL, not a request. Reanalysis-readiness is provenance (attached to the output) plus standards (OME/OME-Zarr) plus a public address.

Where it breaks — The general failure is optimizing for the author's retrieval instead of a stranger's reuse, so the missing context is exactly what the author never noticed they knew — and it differs by platform. A Cell Painting plate deposited without plate maps, perturbation identities, or normalization metadata is pixels no one can profile against. A digital pathology cohort released without scanner, magnification, and stain protocol cannot be pooled with another site, because the reuser cannot model the batch. A light-sheet volume shared without voxel size, channel meaning, or deconvolution provenance is unsegmentable by a third party. A spatial omics dataset missing the registration and panel metadata cannot have its count matrix re-derived. The pattern: completeness is defined by what a stranger lacks, not by what the author remembers — which is why REMBI fields and a public archive, not a good intention, are the mechanism.

References

• Wilkinson et al. (2016). The FAIR Guiding Principles for scientific data management and stewardship. Scientific Data.
• Sarkans et al. (2021). REMBI: Recommended Metadata for Biological Images—enabling reuse of microscopy data in biology. Nature Methods.
• Ellenberg et al. (2024). Enabling global image data sharing in the life sciences (The BioImage Archive). Nature Methods.