Luxar benchmark volumes: sixteen of the seventeen microscopy volumes analysed in the Luxar manuscript (as consumed by the analyses)

Sixteen of the seventeen microscopy volumes analysed in the Luxar manuscript, deposited exactly as the analyses consume them, so the manuscript's measurements can be reproduced without the much larger source recordings. Licence: CC BY-SA 4.0, and why This record contains crops derived from OpenCell (Cho et al. 2022, CZ Biohub SF), published under CC BY-SA 4.0. ShareAlike binds derived products, so the record as a whole is offered under CC BY-SA 4.0 — including volumes whose own licences are more permissive (CC0 or CC BY 4.0) and the author-held volumes. The licence of origin of each volume is given in the table below and per volume in manifest.json; reuse of an individual volume may be governed by that more permissive licence, which you should take from the source rather than from this record. Anything derived from this record as a whole inherits ShareAlike. These are not raw voxel data Every volume is rescaled to [0, 1] float32 by the declarative policy recorded per volume below and in manifest.json. The policies differ, and not every volume is clipped: 10 of 16 are percentile-clipped, and the bounds are not the same across volumes. Some clip only the top (drosophila_gastrulation_t150, flylight_mcfo_tile, h2afva, neuromast_membranes_t050, neuromast_nuclei_t050), while others also clip the low end at the 1st percentile (celegans_t100, opencell_lmnb1_ch0, opencell_lmnb1_ch1, opencell_map4_ch0, opencell_map4_ch1) — which removes dim signal, not just bright outliers. 6 of 16 are not clipped at all: a plain min–max rescale (cells3d_membrane, cells3d_nuclei, kidney_actin, kidney_dapi, organoid_ch0, tribolium). 2 have a constant floor subtracted before normalisation (neuromast_membranes_t050, neuromast_nuclei_t050), and the two floors differ. The floor and the clip are part of the analysed input, not display settings. 4 of 16 volumes are spatial crops of their sources (flylight_mcfo_tile, h2afva, organoid_ch0, tribolium), by the policies listed per volume below; the rest are used whole, with a single timepoint or field of view selected where the source is a series. For h2afva the crop is the densest 512-voxel region of timepoint 252, the last of 253: the source planes are 2048 × 2048, so the window is 512 × 512 laterally and spans all 407 z-planes. flylight_mcfo_tile is the same shape of window: a densest 478-voxel region whose 478 × 478 lateral extent is the window size, spanning all 435 z-planes of its source tile. One benchmark volume is deliberately absent The manuscript also analyses a mouse heart volume (acto3d_heart_nuclei, (597, 174, 960), from the Acto3D project, Takeshita et al. 2024). It is not deposited here: its source distributes the sample data without any stated reuse terms, so we have no permission to redistribute it. Its shape, normalisation policy and array SHA-256 (e7de754dc6767d13770ceb81fad4dfa46ddd9f3485b2c0f8012f2dc8feed078d) are recorded in manifest.json so a reader who obtains it from the Acto3D project can verify they hold the same array the analyses used. Files .npy — canonical. Plain NumPy array, float32, C order; loads with numpy.load and nothing else. 2405 MB in total. .ome.tif — convenience companion. Deflate-compressed OME-TIFF with the voxel size embedded as PhysicalSizeX/Y/Z, so it opens directly in Fiji, napari and QuPath, which cannot read .npy. organoid_ch0 carries no PhysicalSize, because the IDR record for its source states no calibration — better no scale than a fabricated one. manifest.json records, per volume: shape, voxel count, byte count, dtype, value range and mean, source, citation key, modality, marker, voxel size, crop policy and normalisation, plus a SHA-256. That SHA-256 is computed over the raw array bytes (C order, float32) — not over the .npy file, which carries an extra header, and not over the OME-TIFF. Verify with hashlib.sha256(numpy.load(f).tobytes()), not by hashing a file. The export is deterministic. Every array's SHA-256 has been reproduced by at least one fresh load from its own source after the first export: the four author-held volumes on three separate later occasions, the other twelve once, the last of those four days after the first export. No digest has ever changed. volumestructureshape (zyx)voxelsvoxel size (µm)licence of originh2afvaZebrafish embryo nuclei(407, 512, 512)106,692,6081.625, 0.40625, 0.40625author-heldneuromast_membranes_t050Neuromast membranes(84, 580, 576)28,062,7200.25, 0.1083, 0.1083author-heldneuromast_nuclei_t050Neuromast nuclei(84, 580, 576)28,062,7200.25, 0.1083, 0.1083author-helddrosophila_gastrulation_t150Drosophila embryo nuclei(108, 1352, 532)77,680,5121.93, 0.40625, 0.40625author-heldopencell_map4_ch0OpenCell-MAP4 nuclei(51, 600, 600)18,360,0000.5, 0.2, 0.2CC BY-SA 4.0opencell_map4_ch1OpenCell-MAP4 microtubules(51, 600, 600)18,360,0000.5, 0.2, 0.2CC BY-SA 4.0opencell_lmnb1_ch0OpenCell-LMNB1 nuclei(112, 600, 600)40,320,0000.2, 0.2, 0.2CC BY-SA 4.0opencell_lmnb1_ch1OpenCell-LMNB1 nuclear lamina(112, 600, 600)40,320,0000.2, 0.2, 0.2CC BY-SA 4.0kidney_actinKidney actin(16, 512, 512)4,194,3041.25, 1.24, 1.24CC0kidney_dapiKidney nuclei(16, 512, 512)4,194,3041.25, 1.24, 1.24CC0cells3d_membraneCells3D membranes(60, 256, 256)3,932,1600.29, 0.26, 0.26CC0cells3d_nucleiCells3D nuclei(60, 256, 256)3,932,1600.29, 0.26, 0.26CC0organoid_ch0Blastocyst nuclear lamina(236, 275, 271)17,587,900—CC BY 4.0celegans_t100C. elegans embryo nuclei(41, 512, 512)10,747,9040.75, 0.15, 0.15CC BY 4.0flylight_mcfo_tileFly brain neurons(435, 478, 478)99,390,5400.38, 0.19, 0.19CC BY 4.0triboliumTribolium embryo nuclei(991, 104, 965)99,456,7600.381, 0.381, 0.381CC BY 4.0 Sources, licences and per-volume processing zebrafish — Royer lab (Zebrahub), CZ Biohub SF. Licence: author-held. Cite: this work. h2afva: densest 512-voxel region; percentile clip [0, 99.9]. sha256 7f777071a93ace8b5ba13820853de03c2fca3fc5c7a7fb9fa561508ddd76c19e neuromast — Jacobo lab, CZ Biohub SF. Licence: author-held. Cite: Erzberger et al. 2020, Nat Phys 16:949. neuromast_membranes_t050: used whole; constant floor 105.9911880493164 subtracted first, then percentile clip [0, 99.9]. sha256 0ab8dc7a7bd95e627cc7bc8de7381cc69b1e7cc28606ba6bd5aa8a7eafd2fbec neuromast_nuclei_t050: used whole; constant floor 103.88801574707031 subtracted first, then percentile clip [0, 99.9]. sha256 91345eea2899efb49f30b2a97babe9a9bafeddaac1fb297a0692922b4f1b2353 drosophila — Keller lab, HHMI Janelia; SiMView light-sheet, four views registered and fused per timepoint (SiMView pipeline), not deconvolved. Licence: author-held. Cite: Royer et al. 2016, Nat Biotechnol 34:1267. drosophila_gastrulation_t150: used whole; percentile clip [0, 99.9]. sha256 7da821d1ec9f45791f3f41c4856971b3d69db9ab0d6adbc4b19ae51f83cc7d76 opencell_map4 — OpenCell, CZ Biohub SF. Licence: CC BY-SA 4.0. Cite: Cho et al. 2022, Science. Source DOI: 10.1126/science.abi6983. opencell_map4_ch0: used whole; percentile clip [1, 99.5] — note this clips the LOW end too. sha256 6d97997d6d1f23e66fc282e5dfd3c467eacc99ce05218083d439c811f3821740 opencell_map4_ch1: used whole; percentile clip [1, 99.5] — note this clips the LOW end too. sha256 be2f892e4af1b34551f44741d9b04a01dda89b55421e65976ca9093d3d5e47a6 opencell_lmnb1 — OpenCell, CZ Biohub SF. Licence: CC BY-SA 4.0. Cite: Cho et al. 2022, Science. Source DOI: 10.1126/science.abi6983. opencell_lmnb1_ch0: used whole; percentile clip [1, 99.5] — note this clips the LOW end too. sha256 5507fb874ac931bbcd2c8bd1da075bdf3c40e566e13951a03917f5ef49158309 opencell_lmnb1_ch1: used whole; percentile clip [1, 99.5] — note this clips the LOW end too. sha256 e02ab8f91ca679bea270070cf668e3cff010dcb7f5da3c7ea630b6dede628ac2 kidney — G. Buckley, Monash Micro Imaging (scikit-image sample). Licence: CC0. Cite: van der Walt et al. 2014, PeerJ 2:e453. kidney_actin: used whole; min–max rescale to [0, 1] (no clipping). sha256 52eaf4db422d1f1e83ffd75b08bdbc8c609eb67aed2120a760badb8d2a366ff6 kidney_dapi: used whole; min–max rescale to [0, 1] (no clipping). sha256 bb3477775421142de03e273a5d44f85c592c0b3c6c0683ea46106f33afcc0c47 cells3d — Allen Institute for Cell Science (scikit-image cells3d). Licence: CC0. Cite: van der Walt et al. 2014, PeerJ 2:e453. cells3d_membrane: used whole; min–max rescale to [0, 1] (no clipping). sha256 1fa57816214ac8e2be728f60c8afffea86a7e583cf4623329055ec9452c7487f cells3d_nuclei: used whole; min–max rescale to [0, 1] (no clipping). sha256 f786a0399455b9d1f64fe1d71333f28dc9677150754f03ec7e31e53d1f346c1c blastocyst — IDR idr0062 image 6001240, University of Dundee. Licence: CC BY 4.0. Cite: Blin et al. 2019, PLoS Biol 17:e3000388. Source DOI: 10.17867/10000125. organoid_ch0: centred crop along the largest axis, capped at 20,000,000 voxels; min–max rescale to [0, 1] (no clipping). sha256 9770690cac64a50f4e21599d7e9d34fdebc7fc4a348254ef7be3d58a1242905b celegans — Santella, Kovacevic, Bao, Hirsch. Licence: CC BY 4.0. Cite: Murray et al. 2008. Source DOI: 10.5281/zenodo.6460303. celegans_t100: used whole; percentile clip [1, 99.999] — note this clips the LOW end too. sha256 335489af306de6c1017a9d44c4e294de695000098860fafff395c5434c27a893 flylight — Janelia FlyLight Project Team (AWS Open Data). Licence: CC BY 4.0. Cite: Meissner et al. 2023. flylight_mcfo_tile: densest 478-voxel region; percentile clip [0, 99.999]. sha256 2d30a7828b275f74dbcfc779a1919fb624c26368a5b8b0a45e6aa76555a329bb tribolium — D. Barry. Licence: CC BY 4.0. Cite: Preibisch et al. 2014. Source DOI: 10.5281/zenodo.5270323. Licence chain, read this before reuse: that deposit is the GIANI supplementary material, which takes the recording from the Cell Tracking Challenge. The Challenge's own terms require the organisers' permission for public non-Challenge scientific use and forbid redistribution as a competing dataset. The CC BY 4.0 above is the Zenodo deposit's stated licence; it does not extinguish those upstream terms. If you intend to reuse this crop beyond reproducing the measur