Year: 2025

Type: dataset

Abstract: Description This dataset is the U2OS Cell Map Release of Cell Maps for Artificial Intelligence (CM4AI; CM4AI.org), the Functional Genomics Grand Challenge in the NIH Bridge2AI program. It provides an AI-ready, provenance-tracked reconstruction of a multi-scale map of human cell architecture for the U2OS (osteosarcoma) cell line, integrating protein subcellular localization from Human Protein Atlas (HPA) immunofluorescence (IF) imaging with protein-protein interaction (PPI) data from BioPlex affinity purification mass spectrometry (AP-MS), following the methodology of Schaffer, Hu, Qian et al. (Nature, 2025), "Multimodal cell maps as a foundation for structural and functional genomics." The release is delivered as a chain of FAIRSCAPE RO-Crates, each packaging its software, inputs, outputs, and computations with full provenance. It comprises: HPA IF image download for U2OS (cellmaps_imagedownloader) BioPlex U2OS PPI edgelist construction (cellmaps_ppidownloader) ResNet IF image embeddings across two folds (cellmaps_image_embedding) Node2Vec PPI embeddings (cellmaps_ppi_embedding) MUSE co-embedding of the image and PPI modalities across two folds (cellmaps_coembedding) HiDeF multi-scale hierarchy generation (cellmaps_generate_hierarchy) Enrichment-based hierarchy evaluation against CORUM, GO Cellular Component, and HPA (cellmaps_hierarchyeval) Figure-reproduction provenance for the Schaffer et al. (Nature, 2025) publication Data Governance & Ethics Human Subjects: No — data derived from commercially available, de-identified human cell lines FDA Regulated: No Confidentiality Level: Unrestricted Data Governance Committee: Jillian Parker ([email protected]) Ethical Review: Vardit Ravitsky ([email protected]) and Jean-Christophe Bélisle-Pipon ([email protected]) License: CC BY-NC-SA 4.0. Attribution to the copyright holders and authors is required; publications using this data or derived products should cite the related publications and this data collection directly. Completeness The PPI (BioPlex AP-MS) and protein localization (HPA IF imaging) inputs interrogate protein sets that only partially overlap 5,147 proteins are jointly represented in both modalities — roughly half of the detectable U2OS proteome — and only these flow into the co-embedding and downstream hierarchy This release covers the U2OS cell line only Maintenance Plan This is a one-time cell map release and is not planned to be updated. Additional CM4AI cell map releases are anticipated, but they will be issued as separate datasets rather than updates to this one Long-term preservation in the University of Virginia Dataverse, supported by committed institutional funds Intended Use AI-ready datasets to support research in functional genomics Protein complex mapping and multi-scale cell architecture modeling Multimodal representation learning and graph embedding Training, validation, and benchmarking of methods that integrate imaging and interaction data, including interpretable biology-informed ML models Limitations Researchers should be aware of inherent limitations: Scope is limited to the U2OS cell line and may not generalize to other cell types Input IF imaging covers 10,348 proteins (20,660 images) from the HPA; the input PPI network covers 7,543 proteins and 36,842 interactions from BioPlex U2OS AP-MS; only the 5,147 proteins shared by both assays appear in the integrated map AP-MS interactions reflect co-purification and may include indirect interactions; IF localization reflects antibody-accessible epitopes Requires domain expertise for meaningful analysis Prohibited Uses These laboratory data are not to be used in clinical decision-making or in any context involving patient care without appropriate regulatory oversight and approval Potential Sources of Bias Users should be aware of potential biases: Data are derived from a single commercially available osteosarcoma cell line (U2OS) and do not represent biological variation across tissues, donors, or disease states Well-studied proteins, and those with available HPA antibodies or BioPlex bait clones, are over-represented relative to the full U2OS proteome

Source: Libra

DOI: 10.18130/v3/164si1

Institutions: University of California San Diego, University of Virginia, Stanford University

Topic: Bioinformatics and Genomic Networks

Subfield: Molecular Biology

Field: Biochemistry, Genetics and Molecular Biology

Domain: Life Sciences

SDG: Good health and well-being

Open Access status: green

Funder: National Institutes of Health

Award: Bridge2AI: Cell Maps for AI (CM4AI) Data Generation Project

OpenAlex ID: W7167516135

Cell Maps for Artificial Intelligence - U2OS Cell Map