DeepCellMorpho: Cell Painting Embeddings from Foundation Models Pretrained on JUMP-CP

This Zenodo record contains the embeddings and raw data used to generate the figures and quantitative results presented in our Nature Methods manuscript: 'Advancing Morphological Profiling with Self-Supervised Learning: Comparative Insights into Representation Learning of Cellular Perturbations'. The dataset includes representations extracted from two large-scale Cell Painting resources: JUMP Target2-Compounds EU-OPENSCREEN Bioactives dataset (cpg0036) These embeddings are generated using our four SSL models. These embeddings serve as compact, information‑rich descriptors of cellular morphology, enabling different downstream tasks. Embeddings are provided for the four foundation models evaluated in this study, including: MAE ViT-B16 iTPN HiViT-B16 DINO ViT-B16 DINOv2 ViT-B16 For reproducibility, we provide both: Raw embeddings (without_pp) Post-processed embeddings (with_pp) where post-processing consists of the sequential application of: Sphering MAD normalization (Median Absolute Deviation) Harmony batch correction The repository contains the feature representations used to compute all downstream benchmark metrics, including biological target retrieval, perturbation consistency, and cross-source evaluation analyses reported in the manuscript. Data organization Embeddings are stored as serialized Python objects (.pkl) and are organized according to: Dataset (JUMP Target2-Compounds or EU-OPENSCREEN Bioactives dataset (cpg0036)) Feature extraction model Data source/study (e.g., (JUMP-CP- source2, source3...),(CPG0036 - FMP, IMTM, MEDINA, USC)) Cell line (when applicable) Post-processing status (with_pp or without_pp) Reproducibility The purpose of this zenodo record is to enable full reproducibility of the analyses and figures presented in the manuscript. The files can be directly used to recompute benchmark metrics, reproduce published figures, or perform additional downstream analyses. In addition to the embedding files provided in this Zenodo record, the complete benchmarking framework, evaluation pipelines, and analysis code used to generate the manuscript results are available in the DeepCellMorpho repository: https://github.com/johnsonandjohnson/DeepCellMorpho Together, the embeddings hosted here and the DeepCellMorpho codebase provide the resources necessary to fully reproduce the analyses, figures, and benchmarking results reported in the manuscript. Citation If you use these embeddings or derived results, please cite: @article{Abdelmoula2025deepcellmorpho, author = {Abdelmoula, Walid M. and Albuquerque, Tomé and Jaensch, Steffen and Pati, Pushpak and Van Dyck, Michiel and J. Allen, Samantha and Zhang, Litao Zhang and Mansi, Tommaso and Liao, Rui}, title = {Advancing Morphological Profiling with Self-Supervised Learning: Comparative Insights into Representation Learning of Cellular Perturbations}, journal = {Nature Methods}, year = {2026}, volume = {0000}, number = {0}, pages = {}, doi = {}, url = {}, issn = {} } License Please use and redistribute this dataset according to the license (Commons Attribution-NonCommercial 4.0 International (CC BY-NC-SA 4.0)) specified in this Zenodo record.