jump-image-datasets

jump-image-datasets provides packaged JUMP pilot metadata and utilities for downloading image files from metadata tables.

Install

Install from PyPI

pip install jump-image-datasets

Install from PyPI for stable, versioned releases.

Local development with uv

uv venv
uv sync --group test

Editable install

uv pip install -e .

Install from the GitHub repo with pip

pip install "git+https://github.com/WayScience/jump_image_data_downloader.git"

Install from GitHub if you want the latest unreleased changes.

Usage

from jump_image_datasets.jump_pilot import image_downloader, image_metadata

# Load packaged metadata parquet as a DataFrame.
metadata_df = image_metadata.load_metadata()

# Download a small subset.
summary = image_downloader.download_images_with_metadata(
    df=metadata_df.head(10),
    url_column="Metadata_FileUrl",
    default_output_dir="downloaded_jump_pilot_images",
    parallel=True,
    workers=8,
)
print(summary)

For a full runnable example, see docs/download_images_examples.ipynb.

Packaged metadata provenance

This repository ships a packaged metadata table at:

• src/jump_image_datasets/jump_pilot/data/2020_11_04_CPJUMP1_all_plates.parquet

Why this file exists

The file is included so users can immediately load a stable JUMP pilot metadata table (via jump_image_datasets.jump_pilot.image_metadata) without requiring a separate data-fetch or preprocessing step.

How it was created

This parquet was generated from the JUMP Cell Painting Gallery using:

• https://github.com/WayScience/JUMP-single-cell/blob/main/0.download_data/2.download_image_metadata.ipynb

Upstream source pattern used by that notebook:

• s3://cellpainting-gallery/cpg0000-jump-pilot/source_4/workspace/load_data_csv/2020_11_04_CPJUMP1/*/load_data.csv

Transform summary

The generation workflow in 2.download_image_metadata.ipynb:

• Lists all per-plate load_data.csv files for run 2020_11_04_CPJUMP1 (51 files in the captured run) from public S3 (anon=True).
• Reads each plate CSV, appends provenance columns:
  • source_plate (plate ID parsed from path)
  • source_s3_path (full S3 CSV path)
• Concatenates all plate tables into one DataFrame.
• Reshapes channel URL columns from wide to long using melt:
  • URL columns become Metadata_ChannelURLName
  • URL values become Metadata_FileUrl
• Adds normalized channel/stain annotations by mapping URL column names:
  • Metadata_ChannelName: ER, AGP, Mito, DNA, RNA, BF, HZ_BF, LZ_BF
  • Metadata_StainName: corresponding stain labels (or NA for brightfield channels)
• Derives Metadata_Filename from the final path component of Metadata_FileUrl.
• Writes parquet with index=False as data/2020_11_04_CPJUMP1_all_plates.parquet (captured shape: (1495400, 32)).