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

Image-based or morphological profiling is a rapidly expanding field wherein cells are “profiled” by extracting hundreds to thousands of unbiased, quantitative features from images of cells that have been genetically or chemically perturbed. It is the least-expensive high-dimensional profiling technique to date and offers single cell resolution. The Cell Painting assay is the most popular image-based profiling assay; in it, six small-molecule dyes label eight cellular compartments and are imaged in five fluorescence channels (plus brightfield, if desired). Images are segmented and thousands of measurements are then made within the nucleus, cytoplasm, and whole cell, describing quantitative traits such as size and shape, and stain intensity and texture. First published in 2013¹, the standard protocol was updated in 2016 and more recently in 2023².

The field of bioimaging, like most scientific fields, is transitioning towards making data increasingly FAIR (findable, accessible, interoperable, and reusable)³. In this vein, we have created the Cell Painting Gallery, a publicly available collection of Cell Painting datasets, with granular dataset descriptions and comprehensive documentation including access instructions at https://github.com/broadinstitute/cellpainting-gallery. It is hosted by Amazon Web Services (AWS) on the Registry of Open Data (RODA) and anyone can access the data without requiring an AWS account. As of May 2024, the Cell Painting Gallery holds 688 terabytes (TB) of image and associated numerical data. It includes the largest publicly available Cell Painting dataset, in terms of perturbations tested (Joint Undertaking for Morphological Profiling or JUMP), along with many other canonical datasets using Cell Painting and close derivatives (Table 1). Other sources of publicly available Cell Painting datasets include Recursion (RxRx.ai) and Image Data Resource (IDR) (idr.openmicroscopy.org).

Table 1:

Complete datasets available in the Cell Painting Gallery as of publication. Total data size (complete and in-progress datasets) as of May 2024 is 688 TB.

Dataset name	Description	Publication to cite	Objects	Total data size	Image data size	Numerical data size
cpg0000-jump-pilot	300+ compounds and 160+ genes (CRISPR knockout and overexpression) profiled in A549 and U2OS cells, at two time points	Chandrasekaran, S. N. et al. Three million images and morphological profiles of cells treated with matched chemical and genetic perturbations. bioRxiv 2022.01.05.475090 (2022) doi:10.1101/2022.01.05.475090	5.2 M	12.3 TB	6.1 TB	6.1 TB
cpg0001-cellpainting-protocol	300+ compounds profiled in U2OS cells using several different modifications of the Cell Painting protocol. Description of Cell Painting v3	Cimini, B. A. et al. Optimizing the Cell Painting assay for image-based profiling. Nat. Protoc. (2023) doi:10.1038/s41596-023-00840-9	9.6 M	40.3 TB	18.7 TB	21.6 TB
cpg0002-jump-scope	300+ compounds profiled in U2OS using different microscopes and settings	Tromans-Coia, C. et al. Assessing the performance of the Cell Painting assay across different imaging systems. Cytometry A 103, 915–926 (2023)	2.6 M	16.7 TB	12.5 TB	4.2 TB
cpg0003-rosetta	28,000+ genes and compounds profiled in Cell Painting and L1000 gene expression	Haghighi, M., Caicedo, J. C., Cimini, B. A., Carpenter, A. E. & Singh, S. High-dimensional gene expression and morphology profiles of cells across 28,000 genetic and chemical perturbations. Nat. Methods 19, 1550–1557 (2022)	51	8.5 GB	0	8.5 GB
cpg0004-lincs	1,571 compounds across 6 doses in A549 cells	Way, G. P. et al. Morphology and gene expression profiling provide complementary information for mapping cell state. Cell Syst 13, 911–923.e9 (2022)	70.5 M	65.7 TB	61.9 TB	3.8 TB
cpg0010-caie-drugresponse	MCF-7 breast cancer cells treated with 113 small molecules at eight concentrations.	Caie, P. D. et al. High-content phenotypic profiling of drug response signatures across distinct cancer cells. Mol. Cancer Ther. 9, 1913–1926 (2010)	1.1 M	239.2 GB	98.4 GB	140.8 GB
cpg0011-lipocyteprofiler	Variety of lipocytes in different metabolic states and with genetic and drug perturbations	Laber, S. et al. Discovering cellular programs of intrinsic and extrinsic drivers of metabolic traits using LipocyteProfiler. Cell Genomics 3, 100346 (2023)	143 K	1.2 TB	1.2 TB	16 MB
cpg0012-wawer-bioactivecompoundprofiling	30,000 compound dataset in U2OS cells	Wawer, M. J. et al. Toward performance-diverse small-molecule libraries for cell-based phenotypic screening using multiplexed high-dimensional profiling. Proc. Natl. Acad. Sci. U. S. A. 111, 10911–10916 (2014) Bray, M.-A. et al. A dataset of images and morphological profiles of 30 000 small-molecule treatments using the Cell Painting assay. Gigascience 6, 1–5 (2017)	11 M	10.7 TB	3.1 TB	7.6 TB
cpg0015-heterogeneity	2,200+ compounds and 200+ genes profiles in U2OS cells	Rohban, M. H., Abbasi, H. S., Singh, S. & Carpenter, A. E. Capturing single-cell heterogeneity via data fusion improves image-based profiling. Nat. Commun. 10, 2082 (2019)	619	204 GB	0	204 GB
cpg0016-jump	116,000+ compounds and 16,000+ genes (CRISPR knockout and overexpression) profiled in over 1.5 billion U2OS cells.	Chandrasekaran, S. N. et al. JUMP Cell Painting dataset: morphological impact of 136,000 chemical and genetic perturbations. bioRxiv 2023.03.23.534023 (2023) doi:10.1101/2023.03.23.534023	115.3 M	358.4 TB	115.3 TB	243 TB
cpg0017-rohban-pathways	323 genes overexpressed in U2OS cells. Original images re-profiled in 2023	Rohban, M. H. et al. Systematic morphological profiling of human gene and allele function via Cell Painting. Elife 6, (2017)	305 K	321 GB	189 GB	132 GB
cpg0018-singh-seedseq	U2OS cells treated with each of 315 unique shRNA sequences	Singh, S. et al. Morphological Profiles of RNAi-Induced Gene Knockdown Are Highly Reproducible but Dominated by Seed Effects. PLoS One 10, e0131370 (2015)	138 K	247.1 GB	247.1 GB	0
cpg0019-moshkov-deepprofiler	8.3 million single cells from 232 plates, across 488 treatments from 5 public datasets, used for learning representations	Moshkov, N. et al. Learning representations for image-based profiling of perturbations. Nat. Commun. 15, 1594 (2024)	9.3 M	522 GB	482 GB	40 GB
cpg0021-periscope	30 million cells with 20,000 single-gene knockouts in pooled format. A549 cells and HeLa cells in two growth media	Ramezani, M. et al. A genome-wide atlas of human cell morphology. bioRxiv 2023.08.06.552164 (2023) doi:10.1101/2023.08.06.552164	7.1 M	56.0 TB	45.0 TB	11.0 TB
cpg0022-cmqtl	297 iPSC lines	Tegtmeyer, M. et al. High-dimensional phenotyping to define the genetic basis of cellular morphology. Nat. Commun. 15, 347 (2024)	702 K	3.7 TB	2.8 TB	945 GB
cpg0028-kelley-resistance	Bortezomib resistant HCT116 clones	Kelley, M. E. et al. High-content microscopy reveals a morphological signature of bortezomib resistance. Elife 12, (2023)	1 M	4.1 TB	1.9 TB	2.2 TB
cpg0030-gustafsdottir-cellpainting	U2OS cells treated with each of 1600 known bioactive compounds Description of Cell Painting v1	Gustafsdottir, S. M. et al. Multiplex cytological profiling assay to measure diverse cellular states. PLoS One 8, e80999 (2013)	346 K	234 GB	234 GB	.3 GB
cpg0031-caicedo-cmvip	ORF over-expression of 596 alleles of 53 genes in A549 cells	Caicedo, J. C. et al. Cell Painting predicts impact of lung cancer variants. Mol. Biol. Cell 33, ar49 (2022)	553 K	802 GB	605 GB	197 GB

Cell Painting has been widely used for varied biological applications including characterizing gene function (through overexpression, knockout, knockdown, and variant characterization), drug toxicity prediction, mechanism of action identification, and multiple other stages of drug discovery (summarized in⁴). In service of these applications, deep learning methods have been developed that improve on traditional machine learning methods across diverse aspects of high content imaging including image quality, object segmentation, object and image feature extraction, stain prediction, and object and image classification⁵. Curating large, diverse Cell Painting datasets into the Cell Painting Gallery enhances the potential for both novel data mining for biological discovery as well as the development of new machine and deep learning techniques, particularly representation learning and multimodal learning. To that end, we have implemented specific data and metadata guidelines, provided comprehensive download instructions, reprocessed old datasets, converted datasets to next-generation-file-formats, and worked with external organizations to make the Cell Painting Gallery browseable with their infrastructure, all detailed below.

Datasets within Cell Painting Gallery follow specific data and metadata organizational requirements, an important trait of FAIR data. This allows for images and numerical data to easily be downloaded/accessed separately, simplifying application-specific data download. Our data validator analyzes the contents of the Gallery and generates customizable reports of attributes such as data integrity and dataset completion.

To make the Cell Painting Gallery accessible to all researchers regardless of computational or cloud expertise, we provide detailed documentation with instructions for accessing and downloading data from AWS including curating varied example workflows that access data in the gallery.

The field of morphological profiling has undergone major developments in the last 10 years, and like any scientific field, data that was state-of-the-art at publication may be a challenge to use 10 years later. To bring important historical datasets up to date, we have reprocessed several key historical datasets, cpg0012-wawer-bioactivecompoundprofiling, cpg0017-rohban-pathways, and cpg0031-caicedo-cmvip, using current state-of-the-art informatics² so that these modernized datasets now have the same feature set as the ~136,000 chemical and genetic perturbations in cpg0016-jump, the JUMP Cell Painting dataset.

Proprietary image file formats hinder the reuse of both images and workflows. To address this, we have begun converting datasets, such as cpg0004-lincs, to OME-Zarr, a next generation file format optimized for data FAIR-ness and cloud-based usage^6,7. These conversions, performed with tools like Distributed-OMEZarrCreator⁸ also improve interoperability with IDR, soon allowing Cell Painting Gallery data to be browsed through IDR.

The Cell Painting Gallery is continually expanding. Though the majority of the data, apart from JUMP, has been collected at the Broad Institute, we welcome data from other sources and detail in our documentation how to nominate datasets for inclusion. We also welcome community contributions to our documentation, including example workflows and publications that access data in the Gallery. We intend to maintain the Gallery beyond the duration of any individual program or grant.

Data Availability

All data is freely available in the Cell Painting Gallery, hosted by AWS on the Registry of Open Data. An AWS account is not necessary for accessing data. Documentation is provided at https://broadinstitute.github.io/cellpainting-gallery.