fidoptAllen label-free prediction, bright field to lamin B1 fluorescence, 3D
One field of view of live human induced pluripotent stem cells in 3D: the transmitted-light bright-field stack (source) and the fluorescence stack of their mEGFP-tagged lamin B1, the nuclear envelope (target), of one acquisition (Allen Institute for Cell Science, label-free prediction data).
Read before comparing (8)
- Pool. The 80 fields of view the authors list for lamin B1: colonies of live human induced pluripotent stem cells of the Allen Cell Collection, gene-edited to express mEGFP-tagged lamin B1, imaged on 22 plates in May and June 2017 on a Zeiss spinning-disk microscope at 100x. A case is one field of view; fields of one plate share a culture.
- Images. Each field is 624 x 924 pixels of 0.108 µm, 50 to 75 planes 0.29 µm apart, written in micrometres as if millimetres. Thirteen of the 80 files are named '10X', but the three drawn give the same voxel size in their metadata and their nuclei the same size as the others'. The CZI holds 7 or 4 channels; prepare.py takes the bright-field and EGFP channels the authors' file list names and checks them against each file's metadata.
- Values. Both stacks are uint16 camera counts as delivered. The target sits on a camera offset near 400, the nuclear envelope reaching about 1,400 at its 99.9th percentile and up to 4,683; truth.dynamic_range, 1,356, is that percentile pooled over the ten test fields. The bright field's level differs by plate, its median from about 16,000 to 42,000.
- Alignment. Both stacks come from one acquisition, nothing is registered; on the test fields looked at, the lamin outlines fall on the nuclear outlines of the bright field, in plane and along z.
- Mask. The whole field, as the authors score it: the colony fills most of it, and a mask made from the target would give the methods the nuclei's places.
- Baseline. On the whole field of the 10 test cases, at a range of 1,356, a constant at the target's median scores an MAE of 50.5 (39.0 to 68.4), a PSNR of 23.4 dB (21.4 to 25.9) and an SSIM of 0.73 (0.65 to 0.78): the background between the envelopes is most of the field. The bright field as it is, on another scale, scores nothing.
- Outside numbers. The authors report the Pearson correlation of their network's predictions per structure, on their own random quarter of these fields, not this draw; their released models may have seen the test cases.
- Overlap. No other dataset of the corpus holds these images. The Allen WTC-11 single-cell dataset (Viana et al. 2023), from the same institute and pipeline, holds 71 of the 80 fields under the same file names, 17 of these 20 among them: a dataset built from it must leave out all 80, named <plate>_<magnification>_<date>_<scene> as here. Within this one, well F05 of plate 3500000956 gives a test and a development field; they do not overlap.
No method has a result on every case of this dataset yet.
Case by case
Click a score to look at that result: the target image, and over it the method's synthetic image. Underlined scores have their image; for the others you see the case without the result.
About the data
- Test cases
- 10 test fields of view drawn with seed 2026 among the 80 of the authors' file list for lamin B1, then 10 development fields, the last of the same seeded order; one case a field of view. The authors give no test list, only a random three-quarters split in their code.
- Truth
- target image as released the field's mEGFP-lamin B1 fluorescence stack as delivered, taken with the bright-field stack in one acquisition on one spinning-disk microscope, on the same grid, not registered
- Source
- downloads.allencell.org/publication-data/label-free-prediction/index.html
www.allencell.org/terms-of-use.html
doi.org/10.1038/s41592-018-0111-2
doi.org/10.1101/289504
github.com/AllenCellModeling/pytorch_fnet/tree/release_1 - Licence
- Allen Institute for Cell Science terms of use: research and other non-commercial use, copies and derivative works allowed, with its citation policy and its 'freedom to innovate' clause
10 cases for tuning, never tested
3500000958_10X_20170606_3-Scene-04-P22-E06, 3500000924_100X_20170522_4-Scene-03-P41-F04, 3500000956_100X_20170606_2e-Scene-1-P10-E05, 3500000956_100X_20170606_4e-Scene-1-P28-F05, 3500000944_100X_20170530_5-Scene-04-P34-E07, 3500000939_100X_20170526_7-Scene-01-P31-E07, 3500000956_100X_20170606_1e-Scene-1-P1-E04, 3500000949_100X_20170531_6-Scene-4-P15-E05, 3500000876_100X_20170508_3-Scene-06-P26-E06, 3500000938_100X_20170526_5-Scene-7-P63-F07