Cell Painting Gallery: an open resource for image-based profiling

Erin Weisbart, Ankur Kumar, John Arevalo, Anne E Carpenter, Beth A Cimini, Shantanu SinghNature Methods 2024 · DOI · arXiv · OpenAlex

Overview

The Cell Painting Gallery represents a monumental shift in biological data accessibility, hosting over 650 terabytes of image data and numerical profiles. By centralizing diverse datasets and standardizing them into cloud-optimized formats, this open-source repository provides a "FAIR" (Findable, Accessible, Interoperable, and Reusable) foundation designed to accelerate drug discovery and functional genomics through image-based profiling.

Addressing the "Data Silo" Problem

In drug discovery and cell biology, "Cell Painting" is a cost-effective method for profiling cellular responses to chemical or genetic perturbations. By staining cellular compartments and extracting quantitative features, researchers create unique "fingerprints" of cellular states.

However, high-dimensional imaging data is notoriously difficult to manage. It is massive, often stored in proprietary formats, and frequently lacks standardized metadata. Historical datasets often become "dark data"—nearly impossible to align with modern experiments because informatics pipelines have evolved. The Cell Painting Gallery was created to overcome these accessibility and interoperability hurdles by providing a unified, validated infrastructure.

Methodology: Engineering for Accessibility

The authors established a centralized infrastructure on the AWS Registry of Open Data (RODA), focusing on three core pillars of data engineering:

  1. Standardization and Validation: Strict organizational requirements for data and metadata were implemented. A custom data validator tool automatically checks for integrity and completeness, ensuring uploaded data is immediately usable.
  2. Modernizing Legacy Data: To prevent obsolescence, the team reprocessed historical data (some dating back to 2014) using current informatics pipelines. This enables direct comparison between decade-old experiments and contemporary ones.
  3. Cloud-Optimized Formats: Moving away from proprietary formats, the project utilizes OME-Zarr, a cloud-native file format designed for fast, parallel access to massive bioimaging datasets without requiring full local downloads.

A Diverse Catalog of Datasets

The Gallery hosts a wide array of datasets, ranging from pilot studies to massive community-wide efforts.

Table of datasets cpg0000 through cpg0010, including the LINCS dataset (cpg0004).

Key highlights include:

  • JUMP (Joint Undertaking for Morphological Profiling): The "crown jewel" of the collection (cpg0016), representing the largest public Cell Painting dataset with approximately 136,000 chemical and genetic perturbations.
  • cpg0012 (Wawer et al.): A modernized dataset of ~30,000 compounds, now aligned with modern feature sets.
  • cpg0004-lincs: A dataset of 1,571 compounds across multiple doses, recently converted to the OME-Zarr format.
  • Specialized Sets: Includes data on specific cell types (iPSCs), metabolic states (LipocyteProfiler), and pooled CRISPR screens (Periscope).
Table of datasets cpg0011 through cpg0018, featuring the JUMP (cpg0016) and Wawer (cpg0012) datasets.

As of early 2024, the total volume of the gallery stands at 656 Terabytes.

Key Insights: Modernization and Infrastructure

The project highlights that data modernization is as vital as data generation. By reprocessing the 2014 Wawer dataset (cpg0012) to 2023 standards, the authors enabled it to share the same feature set as the massive JUMP dataset, effectively "recycling" historical science for the modern era.

Furthermore, the shift to cloud-native bioimaging via OME-Zarr marks a departure from the traditional "download then analyze" model. Researchers can now stream and analyze specific portions of the 600TB+ dataset. This demonstrates that "FAIRness" requires more than just policy; it requires robust technical infrastructure and validation tools.

Results and Future Directions

The Gallery has successfully achieved unprecedented scale and alignment across disparate datasets. While currently dominated by data from the Broad Institute, the authors are calling for community contributions to diversify the resource. Future work will focus on completing the migration of all datasets to OME-Zarr and deeper integration with the Image Data Resource (IDR) to enhance data browsing and discovery.

Summary updated

BibTeX
@inproceedings{weisbart2024cell,
  author = {Erin Weisbart and Ankur Kumar and John Arevalo and Anne E Carpenter and Beth A Cimini and Shantanu Singh},
  title = {{Cell Painting Gallery: an open resource for image-based profiling}},
  booktitle = {Nature Methods},
  year = {2024},
  doi = {10.1038/s41592-024-02399-z},
}
Figure