Refine CM4AI-DREAM dataset documentation

LICENSE.md

@@ -1,6 +1,6 @@
-# Component licenses and attribution
+# License notices and attribution
 
-This repository combines two training releases and retains their separate original license notices:
+The CM4AI-DREAM training data retain their original license notices:
 
 - [HPA-train_v1.0/LICENSE.txt](HPA-train_v1.0/LICENSE.txt)
 - [KOLF-train_v1.0/LICENSE.txt](KOLF-train_v1.0/LICENSE.txt)

README.md

@@ -16,21 +16,15 @@ viewer: false
 
 # CM4AI-DREAM training data
 
-**Release status: preparation in progress. The complete training data have not yet been uploaded to this repository.** The layout and download instructions below describe the intended complete release.
-
 CM4AI-DREAM provides single-cell fluorescence microscopy training data for conditional protein-stain generation. Morphology stains, cell-line information, target protein identifiers and protein sequences provide context for the observed protein stain.
 
-This repository brings the HPA and KOLF training releases together while preserving their original folder structures. It contains training data; challenge evaluation data are not included.
+The HPA images derive from Human Protein Atlas version 25 and were cropped to single cells, resized and converted to 8-bit images.
 
-## Release contents
+The KOLF images were generated by the Cell Maps for Artificial Intelligence (CM4AI) project in KOLF2.1J human induced pluripotent stem cells. They were background-subtracted, normalized per well, converted to 8-bit images and cropped to 512 × 512 pixel single-cell images.
 
-| Folder | TAR archives | Other files | Total files | Size (decimal GB) |
-| --- | ---: | --- | ---: | ---: |
-| `HPA-train_v1.0/` | 15,674 | `hpa_train_manifest.csv.gz`, `LICENSE.txt` | 15,676 | 609.224 |
-| `KOLF-train_v1.0/` | 495 | `kolf_train_manifest.csv.gz`, `LICENSE.txt` | 497 | 80.383 |
-| **Combined data release** | **16,169** | Two manifests and two licenses | **16,173** | **689.607** |
+Protein sequences come from Ensembl release 109.
 
-The data-release size is **689,607,128,597 bytes**; repository documentation is additional. The largest release file is approximately 285 MB.
+## Release contents
 
 ```text
 HPA-train_v1.0/
@@ -47,11 +41,9 @@ KOLF-train_v1.0/
     HPA00/ ... HPA07/
 ```
 
-Keep the two release folders separate. TAR basenames and `LICENSE.txt` names overlap between the releases, so flattening or merging the folders can overwrite files.
-
 ## Download
 
-After the release becomes available, download it with the Hugging Face CLI:
+Download the dataset with the Hugging Face CLI:
 
 ```bash
 hf download LundbergLab/CM4AI-DREAM --repo-type dataset --local-dir CM4AI-DREAM
@@ -64,11 +56,9 @@ hf download LundbergLab/CM4AI-DREAM --repo-type dataset \
   --include 'KOLF-train_v1.0/**' --local-dir CM4AI-DREAM
 ```
 
-For a private or gated release, first authenticate with an account that has access using `hf auth login`. For reproducible work, add `--revision` with the published release tag or commit hash. Allow space for both the downloaded archives and any images you choose to extract.
-
 ## Manifest and archive format
 
-Each compressed CSV manifest has one row per cell. TAR archives group the corresponding image files and protein sequences. Both manifests use the following columns:
+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 |
 | --- | --- |
@@ -88,7 +78,7 @@ Each compressed CSV manifest has one row per cell. TAR archives group the corres
 | `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. These are the original release archives; they are not advertised as Hugging Face `load_dataset()` or WebDataset-compatible shards. The automatic dataset viewer is disabled.
+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:
 
@@ -116,21 +106,11 @@ with tarfile.open(matches[0], "r:") as archive:
 print(row["gene_name"], row["cell_line"], image.size)
 ```
 
-The image-reading example requires Pillow; the other imports are part of Python's standard library.
-
-## Sources and preprocessing
-
-The HPA component derives from Human Protein Atlas version 25. Its release license describes single-cell cropping, resizing and conversion to 8-bit images.
-
-The KOLF component was generated by the Cell Maps for Artificial Intelligence (CM4AI) project in KOLF2.1J human induced pluripotent stem cells. Its release license describes background subtraction, normalization per well, conversion to 8-bit images and cropping to 512 × 512 pixel single-cell images.
-
-Both release licenses identify Ensembl release 109 as the source of protein sequences. The original release files will be preserved without repackaging or additional image transformations.
-
 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. The root [license note](LICENSE.md) explains how the component notices relate to this combined repository.
+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: