File size: 4,971 Bytes
fb895b0 c435668 ac034fd fb895b0 ac034fd fb895b0 4b030c0 fb895b0 4b030c0 fb895b0 4b030c0 fb895b0 4b030c0 fb895b0 4b030c0 fb895b0 4b030c0 fb895b0 c435668 fb895b0 c435668 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 | ---
pretty_name: CM4AI-DREAM Training Data
license: other
license_name: component-specific-attribution-licenses
license_link: LICENSE.md
tags:
- biology
- bioimaging
- microscopy
- protein-localization
- conditional-image-generation
- cm4ai
- dream-challenge
viewer: false
---
# CM4AI-DREAM training data
CM4AI-DREAM provides fluorescence microscopy training data for machine learning (ML) model training, such as development of generative ML tools that simulate protein staining.
Reference marker staining images (DNA, Microtubules, ER), cell-line information, target protein identifiers and protein sequences provide context for the observed protein stain.
The HPA images have been derived from Human Protein Atlas version 25. The KOLF images were generated by the Cell Maps for Artificial Intelligence (CM4AI) project in KOLF2.1J human induced pluripotent stem cells. Each image is a 512 × 512 pixel crop centered on a target cell, spanning approximately 48.2 × 48.2 μm (0.0942 μm per pixel). Crops can also contain neighboring cells.
Protein sequences come from Ensembl release 109.
## Release contents
```text
HPA-train_v1.0/
LICENSE.txt
hpa_train_manifest.csv.gz
tars/
CAB00/ ... CAB08/
HPA00/ ... HPA07/
KOLF-train_v1.0/
LICENSE.txt
kolf_train_manifest.csv.gz
tars/
CAB07/ CAB08/
HPA00/ ... HPA07/
```
## Download
Download the dataset with the Hugging Face CLI:
```bash
hf download LundbergLab/CM4AI-DREAM --repo-type dataset --local-dir CM4AI-DREAM
```
To download one component:
```bash
hf download LundbergLab/CM4AI-DREAM --repo-type dataset \
--include 'KOLF-train_v1.0/**' --local-dir CM4AI-DREAM
```
## Manifest and archive format
Each compressed CSV manifest has one row per cell. TAR archives group the corresponding image files and protein sequences. The manifests use the following columns:
| Column | Contents |
| --- | --- |
| `image_key` | Cell-image identifier |
| `antibody_id` | Antibody identifier |
| `ensembl_protein_id` | Ensembl protein identifier |
| `ensembl_gene_id` | Ensembl gene identifier |
| `gene_name` | Gene name |
| `cell_line` | Cell-line name |
| `locations` | Supplied localization annotation; may be empty |
| `fov_key` | Field-of-view identifier |
| `tar_file` | TAR basename |
| `protein_path` | Protein-stain PNG member path inside the TAR |
| `nucleus_path` | Nucleus-stain PNG member path inside the TAR |
| `microtubules_path` | Microtubule-stain PNG member path inside the TAR |
| `er_path` | Endoplasmic-reticulum-stain PNG member path inside the TAR |
| `mask_path` | Mask PNG member path inside the TAR |
| `sequence_path` | Protein sequence member path inside the TAR |
Resolve `tar_file` within the same release's `tars/` subdirectories. The other path columns refer to members **inside** that archive, rather than standalone repository files.
For example, inspect a row and open its protein image without extracting a whole archive:
```python
import csv
import gzip
import io
from pathlib import Path
import tarfile
from PIL import Image
release = Path("CM4AI-DREAM/KOLF-train_v1.0")
with gzip.open(release / "kolf_train_manifest.csv.gz", "rt", newline="") as stream:
row = next(csv.DictReader(stream))
matches = list((release / "tars").glob("*/" + row["tar_file"]))
if len(matches) != 1:
raise RuntimeError("Download the matching TAR archive for this row first")
with tarfile.open(matches[0], "r:") as archive:
member = archive.extractfile(row["protein_path"])
if member is None:
raise RuntimeError("Protein image is missing from the archive")
image = Image.open(io.BytesIO(member.read()))
image.load()
print(row["gene_name"], row["cell_line"], image.size)
```
Cells from the same field of view or acquisition can share biological and technical variation. Localization annotations provide context and should not be treated as independently verified localization ground truth for every cell.
## Licensing and attribution
The image and annotation licenses specify **Creative Commons Attribution 4.0 International (CC BY 4.0)**, with different source attributions. Read the original [HPA license](HPA-train_v1.0/LICENSE.txt) and [KOLF license](KOLF-train_v1.0/LICENSE.txt) and retain the appropriate attributions. See the root [license note](LICENSE.md) for a summary.
Please cite the underlying sources listed in those notices:
- Thul PJ et al. *A subcellular map of the human proteome.* Science 356, eaal3321 (2017). DOI: [10.1126/science.aal3321](https://doi.org/10.1126/science.aal3321)
- Clark T et al. *Cell Maps for Artificial Intelligence: AI-Ready Maps of Human Cell Architecture from Disease-Relevant Cell Lines.* bioRxiv 2024.05.21.589311 (2024). DOI: [10.1101/2024.05.21.589311](https://doi.org/10.1101/2024.05.21.589311)
## Maintainer
Published by the Lundberg Lab for the [CM4AI-DREAM challenge](https://synapse.org/CM4AI_DREAM). Report dataset questions through this repository's Community tab.
|