U2OS FKHR-GFP translocation (BBBC013)

A cytoplasm-to-nucleus translocation assay from the Broad Bioimage Benchmark Collection (BBBC013v1), contributed by Ilya Ravkin. One 96-well plate of human osteosarcoma (U2OS) cells stably expressing a Forkhead (FKHR-EGFP) fusion protein: in proliferating cells FKHR sits in the cytoplasm, and when nuclear export is blocked by inhibiting PI3K/PKB signalling with wortmannin or LY294002 for 1 h it accumulates in the nucleus. Nuclei are stained with DRAQ. Images were acquired at BioImage on an IN Cell Analyzer 3000, one field per well, one image per channel. The source ships the images as a flat folder plus three platemap text files; here the platemap is attached to every image as file metadata and as an image-level label, so wells can be filtered by drug, dose and control role in the browser.

Contents

Images	192 (96 wells x 2 channels, 640 x 640, 8-bit grayscale BMP, byte-identical to the source BBBC013_v1_images_bmp.zip)
Annotations	192 image-level labels (one per file, the well's control role); no region annotations, the source has no outlines
Classes	3: negative-control (16 files, 8 wells), positive-control (16 files, 8 wells), dose (160 files, 80 wells)
Dataset	translocation-bbbc013-v1
Split	None. One plate: rows A-D are four replicate 9-point wortmannin series (nM), rows E-H four replicate 9-point LY294002 series (uM)
Imaging	Fluorescence; channel 1 FKHR-GFP, channel 2 DNA (DRAQ)

Plate layout

Column	Rows A-D (wortmannin half)	Rows E-H (LY294002 half)
1	negative control, no drug	positive control, 150 nM wortmannin
2	0 dose	0 dose
3-11	wortmannin 0.98, 1.95, 3.91, 7.81, 15.63, 31.25, 62.5, 125, 250 nM	LY294002 0.31, 0.63, 1.25, 2.5, 5, 10, 20, 40, 80 uM
12	positive control, 150 nM wortmannin	negative control, no drug

The zero-dose wells in column 2 are labelled dose with dose = 0 (the source treats them as the first point of each dose curve, not as controls).

Files, metadata and channel pairing

Files keep their source names, Channel<c>-<well index 1-96>-<row>-<column>.BMP, under images/<control role>/. The two channels of a well are two separate files (no merged TIFF was made) with the same name apart from the Channel1/Channel2 prefix, in the same folder; each file's pair_file and pair_path metadata point at its partner.

Metadata keys on every file: well, row, column, well_index, channel, stain, plate_half (wortmannin or ly294002, by row), control_role, treatment (drug actually in the well: wortmannin, ly294002 or none), dose, dose_unit (nM or uM, empty for no-drug wells), platemap_dose (the raw value from BBBC013_v1_platemap_all.txt), pair_file, pair_path.

Using it

Browse any image to see its label and metadata. To download everything (free account required):

• ZIP: use the Download button on this project.
• Python SDK: pip install datatorch, then:

datatorch login
datatorch pull datatorch/translocation-bbbc013

Annotations export in COCO format through the project's export schema; the labels are per image, not per region, so exports carry categories and image metadata but no boxes or segmentations.

Conversion notes

• Images are the unmodified files from BBBC013_v1_images_bmp.zip (32,404,692 bytes). They were not re-encoded. Every file was checked by parsing its BMP header and pixel data in Python: all 192 are 640 x 640, 8-bit, uncompressed, with a grayscale palette, and the declared sizes match the bytes on disk.
• Checks: 192 of 192 files decode; no byte-identical duplicates (SHA-256); every file sits in exactly one class folder; every well has exactly one file per channel; the well index in each filename equals the row-major well number.
• The platemap text files on the source page contain no well identifiers, only 96 dose values in row-major well order (A01 ... H12), and the first value is glued to the header line by a stray carriage return. They were decoded by position. Doses in the all and wortmannin/ly294002 files agree.
• Control roles are not stated per well on the source page or in the platemap files. They were taken from the per-well metadata CSVs in the source's BBBC013_reproduce_logan.zip (Logan and Carpenter, J. Biomol. Screening, 2010), which mark column 1 of rows A-D and column 12 of rows E-H as negative controls and column 12 of rows A-D and column 1 of rows E-H as positive controls (150 nM wortmannin).
• Discrepancy: BBBC013_v1_platemap_all.txt and _ly294002.txt give wells E01, F01, G01 and H01 the value 80 (the top LY294002 dose), while the Logan metadata (file name ends in _Corrected_EFGH01_Controls) records them as 150 nM wortmannin positive controls. This dataset follows the corrected Logan metadata for control_role, treatment and dose, and keeps the raw platemap value in platemap_dose.
• A file named translocation_doses_and_controls.csv is sometimes searched for alongside this set; it is not on the source page and is not part of this dataset.
• The native FRM images (BBBC013_v1_images_frm.zip) were not mirrored; this project uses the 8-bit BMP release only.

License and citation

CC BY 3.0 Unported. The BBBC013 images are licensed under a Creative Commons Attribution 3.0 Unported License by Ilya Ravkin (license badge and statement on the source page). If you use this data, please cite:

We used image set BBBC013v1 provided by Ilya Ravkin, available from the Broad Bioimage Benchmark Collection [Ljosa et al., Nature Methods, 2012].

Source: BBBC013v1, Broad Bioimage Benchmark Collection. Original context: Ilya Ravkin's Invitation to Participate in the Comparison of Image Analysis Algorithms for Intracellular Screening.

Related

• More microscopy datasets on DataTorch
• All public datasets