CM4AI d4d

Datasheet for Dataset - Human Readable Format

🎯

Motivation

Why was the dataset created?

DescriptionID
Deliver machine-readable hierarchical maps of cell architecture as AI-Ready data from multimodal interrogation of disease-relevant cell lines to enable transformative biomedical AI research. CM4AI produces integrated cell maps from spatial proteomics, protein-protein interactions, and genetic perturbations using state-of-the-art mass spectrometry, cell imaging, and CRISPR technologies.
cm4ai:purpose:1
Address the grand challenge of interpretable genotype-phenotype learning in genomics and precision medicine. Machine learning models are often "black boxes" predicting phenotypes from genotypes without understanding the mechanisms. CM4AI enables "visible" machine learning systems informed by multi-scale cell and tissue architecture, allowing AI tools to interrogate how protein assemblies in the cell affect cell-level phenotype predictions.
cm4ai:purpose:2
Establish standards, best practices, and guidelines for ethical AI-readiness in biomedical data. This includes implementing FAIR principles, computing machine-readable provenance graphs, characterizing and validating all datasets with JSON-Schema mini-data-dictionaries, and mapping data elements to public ontology vocabularies where appropriate.
cm4ai:purpose:3
Provide multimodal cell maps as a foundation for structural and functional genomics, enabling integrative structure modeling of protein assemblies identified via the MuSIC pipeline. Demonstrated in peer-reviewed publication in Nature (Schaffer, Hu et al., April 2025, doi:10.1038/s41586-025-08878-3) integrating IF imaging, AP-MS, SEC-MS, and structure modeling.
cm4ai:purpose:4
  • ID
    cm4ai:funder:1
    Description
    National Institutes of Health Common Fund Bridge2AI Program. Funded through NIH grant 1OT2OD032742-01 (Bridge2AI Functional Genomics) and 5U54HG012513-02 (Bridge2AI Bridge Center), administered by NIH Office of the Director. Opportunity Number: OTA-21-008. Project dates: September 1, 2022 to August 31, 2026. FY 2025 funding: $5,289,382 (Direct: $4,632,095, Indirect: $657,287). Additional funding from the Frederick Thomas Fund of the University of Virginia.
📊

Composition

What do the instances represent?

DescriptionID
MDA-MB-468: Triple negative breast cancer cell line (RRID:CVCL_0419) established from a metastatic site pleural effusion of a 51-year-old black female with a metastatic mammary adenocarcinoma, available from ATCC. This cell line has been extensively used to study triple-negative breast cancer and is well characterized with transcriptomic, mutational profile, and whole-genome sequencing data available. Cells are analyzed under three conditions: untreated, paclitaxel-treated, and vorinostat-treated.
cm4ai:instance:1
KOLF2.1J: Human induced pluripotent stem cell (iPSC) line (RRID:CVCL_B5P3) derived from a healthy male Northern European donor, available from the Human Induced Pluripotent Stem Cells Initiative (HipSci) resource. Available for access by non-for-profit organizations via a simple MTA. Analyzed in undifferentiated state and after differentiation into neurons, neural progenitor cells (NPCs), and cardiomyocytes.
cm4ai:instance:2
100 Chromatin Regulators: Near-comprehensive set of chromatin regulators encoded by the human genome analyzed via AP-MS, SEC-MS, IF imaging, and CRISPR perturbation screens across different cell states and treatment conditions. 17 genes endogenously tagged in MDA-MB-468 with AP-MS data under three conditions, with 34 additional genes in process. SEC-MS identified 72/100 chromatin modifiers, with 52 being integral components of protein complexes.
cm4ai:instance:3
100 Metabolic Enzymes: Set of metabolic enzymes involved in cancer, neuropsychiatric, and cardiac disorders analyzed via multimodal interrogation including mass spectrometry, imaging, and perturbation screens.
cm4ai:instance:4
DescriptionID
MDA-MB-468 Untreated: MDA-MB-468 breast cancer cells in control/untreated conditioncm4ai:subpop:1
MDA-MB-468 Paclitaxel-Treated: MDA-MB-468 breast cancer cells treated with paclitaxel chemotherapycm4ai:subpop:2
MDA-MB-468 Vorinostat-Treated: MDA-MB-468 breast cancer cells treated with vorinostat chemotherapycm4ai:subpop:3
KOLF2.1J Undifferentiated iPSCs: KOLF2.1J induced pluripotent stem cells in undifferentiated/naive statecm4ai:subpop:4
KOLF2.1J iPSC-Derived Neurons: KOLF2.1J iPSCs differentiated into neuronscm4ai:subpop:5
KOLF2.1J iPSC-Derived Neural Progenitor Cells (NPCs): KOLF2.1J iPSCs differentiated into neural progenitor cellscm4ai:subpop:6
KOLF2.1J iPSC-Derived Cardiomyocytes: KOLF2.1J iPSCs differentiated into cardiomyocytescm4ai:subpop:7
Access UrlsDescriptionID
https://doi.org/10.18130/V3/DXWOS5, https://doi.org/10.18130/V3/B35XWX, https://doi.org/10.18130/V3/F3TD5R, https://doi.org/10.18130/V3/K7TGEM, https://www.cm4ai.org
RO-Crate Packages with Provenance: All CM4AI output data packaged as Research Object Crate (RO-Crate) packages containing datasets, metadata, provenance graphs, and software (or resolvable references). RO-Crates assigned persistent globally unique identifiers (ARK scheme, DOIs planned for publishable work) that resolve to machine- and human-readable landing pages with metadata in JSON-LD using Schema.org and EVI vocabularies.
cm4ai:format:1
MassIVE Repository (human iPSCs - SEC-MS), MassIVE Repository (human cancer cells - SEC-MS)
Mass Spectrometry Data in MassIVE: Mass spectrometry data deposited to MassIVE Repository (Proteomics community-supported repository). Separate depositions for human iPSC data and human cancer cell data (SEC-MS for KOLF2.1J iPSCs and MDA-MB-468 cancer cells). Data will be uploaded to PRIDE when available.
cm4ai:format:2
NCBI BioProject, Sequence Read Archive (SRA)
Sequence Data in NCBI SRA: Raw sequence data from CRISPR perturbation screens deposited to NCBI BioProject/Sequence Read Archive (SRA). Genome-scale CRISPRi perturbation cell atlas raw sequences and processed data available.
cm4ai:format:3
https://www.ndexbio.org
Hierarchical Cell Maps in NDEx: Cell maps shared via Network Data Exchange (NDEx) for visualization and access. Maps can be visualized in web browser or accessed via tools such as Cytoscape, HiView, and Python ndex2 library.
cm4ai:format:4
https://doi.org/10.18130/V3/DXWOS5, LibraData University of Virginia
University of Virginia Dataverse: Archived RO-Crates available in University of Virginia's LibraData data archive (instance of Harvard's Dataverse, an NIH-approved generalist repository). Long-term preservation supported by committed institutional funds. Quarterly updates through November 2026.
cm4ai:format:5
🔍

Collection Process

How was the data acquired?

https://doi.org/10.18130/V3/DXWOS5
CM4AI
Cell Maps for Artificial Intelligence (CM4AI)
CM4AI is the Functional Genomics Data Generation Project in the U.S. National Institutes of Health's (NIH) Bridge to Artificial Intelligence (Bridge2AI) program. Its overarching mission is to produce ethical, AI-ready datasets of cell architecture, inferred from multimodal data collected for human cell lines, to enable transformative biomedical AI research. The project delivers machine-readable hierarchical maps of cell architecture as AI-Ready data produced from multimodal interrogation of 100 chromatin modifiers and 100 metabolic enzymes involved in cancer, neuropsychiatric, and cardiac disorders in disease-relevant cell lines under perturbed and unperturbed conditions. Data streams include immunofluorescence (IF) subcellular microscopy for spatial proteomics, affinity purification mass spectroscopy (AP-MS) and size exclusion mass spectroscopy (SEC-MS) for protein-protein interaction (PPI) data, and single-cell CRISPR-Cas perturbation screens by cell type. Input data streams are integrated via the Multi-Scale Integrated Cell (MuSIC) software pipeline employing deep learning models and community detection algorithms, and output cell maps are packaged with provenance graphs and rich metadata as AI-Ready datasets in RO-Crate format using the FAIRSCAPE framework. A Nature publication (Schaffer, Hu et al., April 2025) demonstrates multimodal cell maps as a foundation for structural and functional genomics, integrating IF imaging, AP-MS, and SEC-MS with integrative structure modeling to produce multimodal cell maps of MDA-MB-468 breast cancer cells and KOLF2.1J iPSCs.
https://www.cm4ai.org
en
  • Cell Maps
  • Artificial Intelligence
  • AI-Ready Data
  • Bridge2AI
  • Functional Genomics
  • Protein-Protein Interactions
  • Spatial Proteomics
  • CRISPR Perturbation
  • Hierarchical Cell Maps
  • MDA-MB-468
  • KOLF2.1J
  • iPSC
  • Breast Cancer
  • Chromatin Modifiers
  • Metabolic Enzymes
  • Immunofluorescence
  • Mass Spectrometry
  • AP-MS
  • SEC-MS
  • Perturb-Seq
  • FAIR Principles
  • RO-Crate
  • FAIRSCAPE
  • Visible Neural Networks
  • Deep Learning
  • Integrative Structure Modeling
DescriptionID
Address the limitation that machine learning models in genomics and precision medicine are typically difficult-to-interpret "black boxes" by providing hierarchical cell maps that enable visible machine learning systems built directly on knowledge maps of cell and tissue architecture.
cm4ai:gap:1
Provide fully provenanced, ethically validated, and FAIR-compliant AI-ready datasets with machine-readable provenance graphs, complete schemas, validation procedures, and data sheets that can be reliably processed by AI applications with full explainability.
cm4ai:gap:2
Create integrated datasets combining protein localization (spatial proteomics), protein-protein interactions (AP-MS and SEC-MS), and transcriptional states (CRISPR perturbation screens) at multiple scales, enabling complex multi-modal AI analyses not feasible with single data types.
cm4ai:gap:3
Bridge the gap between protein interaction networks and structural biology by enabling integrative structure modeling of protein communities identified from multimodal cell maps, combining PDB, AlphaFoldDB, crosslinking MS, and sequence disorder predictions.
cm4ai:gap:4
RoleNameORCIDAffiliation
Contributorcm4ai:creator:1-
Contributorcm4ai:creator:2-
Contributorcm4ai:creator:3-
Contributorcm4ai:creator:4-
Contributorcm4ai:creator:5-
Contributorcm4ai:creator:6-
Contributorcm4ai:creator:7-
Contributorcm4ai:creator:8-
Contributorcm4ai:creator:9-
Contributorcm4ai:creator:10-
Contributorcm4ai:creator:11-
Contributorcm4ai:creator:12-
Contributorcm4ai:creator:13-
Contributorcm4ai:creator:14-
Contributorcm4ai:creator:15-
Contributorcm4ai:creator:16-
Contributorcm4ai:creator:17-
Contributorcm4ai:creator:18-
Contributorcm4ai:creator:19-
Contributorcm4ai:creator:20-
Contributorcm4ai:creator:21-
Contributorcm4ai:creator:22-
Contributorcm4ai:creator:23-
Contributorcm4ai:creator:24-
DescriptionID
Spatial Proteomics IF Images - MDA-MB-468: Immunofluorescence-based staining (ICC-IF) and confocal microscopy images displaying spatial localization of proteins of interest in MDA-MB-468 breast cancer cells under three conditions: untreated, paclitaxel-treated, and vorinostat-treated. March 2025 Beta release included 563 proteins; June 2025 Beta V2.1 revision includes 464 proteins with RGB images. Nuclei stained with DAPI (blue channel), endoplasmic reticulum with calreticulin antibody (yellow channel), microtubules with tubulin antibody (red channel), and antibody against protein of interest (green channel). Generated by Lundberg Lab at Stanford University using automated fixation and permeabilization protocols.
cm4ai:subset:1
Protein-Protein Interaction AP-MS Data: Affinity purification mass spectrometry (AP-MS) data on endogenously tagged cell lines mapping protein-protein interactions of chromatin regulators. 17 genes endogenously tagged in MDA-MB-468 with data acquired under three conditions (untreated, paclitaxel, vorinostat). Orthogonal approach to SEC-MS for PPI mapping.
cm4ai:subset:2
Protein-Protein Interaction SEC-MS Data: Size exclusion chromatography coupled to mass spectrometry (SEC-MS) for proteome-wide complex/interaction mapping. Performed on MDA-MB-468 cells under three conditions (untreated, paclitaxel, vorinostat) and on KOLF2.1J iPSCs and derivatives (undifferentiated, NPCs, neurons, cardiomyocytes). Detected PPI profiles of over 1,000 complexes in MDA-MB-468 cells and over 700 protein complexes in iPSCs and differentiated neurons. Identified 72/100 chromatin modifiers with 52 being integral components of protein complexes. October 2025 release adds SEC-MS for MDA-MB-468 breast cancer cells.
cm4ai:subset:3
CRISPR Perturbation Cell Atlas: Genome-scale CRISPRi perturbation cell atlas in undifferentiated KOLF2.1J human induced pluripotent stem cells (hiPSCs) mapping transcriptional and fitness phenotypes associated with 11,739 targeted genes. Single-cell CRISPR screens performed using 10x Genomics 3'HT kit with CRISPR lentiviral library targeting 100 chromatin factors with 6 guide RNAs per gene. Screens conducted in MDA-MB-468 cells under 3 conditions (no treatment, paclitaxel, vorinostat) and KOLF2.1J iPSC in undifferentiated state. October 2025 release adds Perturb-seq data for MDA-MB-468 breast cancer cells. Includes raw sequence data and processed cell atlas data.
cm4ai:subset:4
Hierarchical Cell Maps via MuSIC: Integrated hierarchical cell maps produced by Multi-Scale Integrated Cell (MuSIC) pipeline from fusion of protein localization (IF images) and protein-protein interaction data (AP-MS and SEC-MS). Cell maps are hierarchical directed acyclic graphs (DAG) where each node represents an assembly of proteins in proximity at a given scale, spanning from large assemblies representing cell compartments to small assemblies of protein complexes. Maps contain 10 layers of depth representing communities at multiple resolutions, with communities at smaller distance nested inside larger communities. Nature publication (Schaffer, Hu et al., April 2025) presents first fully integrated multimodal cell maps.
cm4ai:subset:5
  1. ID
    cm4ai:sampling:1
    Description
    Purposive selection of two disease-relevant cell lines: MDA-MB-468 triple negative breast cancer cell line for cancer research, and KOLF2.1J iPSCs for neuropsychiatric and cardiac disorder research. Both cell lines ethically sourced and well-characterized in the literature. Selection criteria: (1) MDA-MB-468 chosen for triple-negative breast cancer research applications; (2) KOLF2.1J chosen as reference iPSC line for large-scale collaborative studies; (3) both have extensive existing characterization data and are commercially available and ethically sourced.
    Sample
    False
    Random Sampling
    False
    Representative Sample
    False
    Strategies
    • Selection of commercially available, ethically sourced cell lines
    • MDA-MB-468 chosen for triple-negative breast cancer research applications
    • KOLF2.1J chosen as reference iPSC line for large-scale collaborative studies
    • Both cell lines have extensive existing characterization data
DescriptionID
Immunofluorescence Spatial Proteomics Imaging: Automated fixation and permeabilization protocols using pipetting robot for MDA-MB-468 and KOLF2.1J cell lines. Immunofluorescence-based staining (ICC-IF) with confocal microscopy to capture spatial subcellular organization. Completed spatial proteomics mapping of 100 chromatin regulators in MDA-MB-468 cells under three conditions (untreated, paclitaxel, vorinostat), with 500 additional proteins pending from genetic perturbations and PPI results. Antibodies from Human Protein Atlas resource. Generated by Lundberg Lab at Stanford University.
cm4ai:collection:1
Affinity Purification Mass Spectrometry: Endogenous tagging of genes in cell lines followed by affinity purification mass spectrometry (AP-MS) to map protein-protein interactions. 17 genes endogenously tagged in MDA-MB-468 with AP-MS data acquired under three conditions (untreated, paclitaxel, vorinostat). 34 additional genes currently in tagging process. Orthogonal approach to SEC-MS for comprehensive PPI mapping.
cm4ai:collection:2
Size Exclusion Chromatography Mass Spectrometry: Size exclusion chromatography coupled to mass spectrometry (SEC-MS) for proteome-wide complex/interaction mapping. Performed in Krogan Laboratory at UCSF. Conducted on MDA-MB-468 cells under three conditions and on KOLF2.1J iPSCs and derivatives (undifferentiated, NPCs, neurons, cardiomyocytes). Enabled detection of over 1,000 protein complexes in MDA-MB-468 cells and over 700 complexes in iPSCs, with thousands of proteins exhibiting differential elution profiles between control and treated cells.
cm4ai:collection:3
CRISPR Perturbation Screens: Single-cell CRISPR screens using CRISPR lentiviral library targeting 100 chromatin factors with 6 guide RNAs per gene. Generated and characterized MDA-MB-468 and KOLF2.1J CRISPR lines expressing inducible dCas9. Screens performed in MDA-MB-468 cells under 3 conditions (no treatment, paclitaxel, vorinostat) and in undifferentiated KOLF2.1J iPSCs using 10x Genomics 3'HT kit. Genome-scale screens mapping transcriptional and fitness phenotypes for 11,739 targeted genes.
cm4ai:collection:4
DescriptionID
Confocal Microscopy for Subcellular Imaging: High-resolution confocal microscopy of immunofluorescence-stained cells capturing four channels: DAPI (nuclei, blue), calreticulin antibody (ER, yellow), tubulin antibody (microtubules, red), and antibody against protein of interest (green). Images processed using Human Protein Atlas deep learning model to reduce dimensionality, producing image embeddings containing information about protein localization.
cm4ai:acquisition:1
Mass Spectrometry for Protein Interactions: State-of-the-art mass spectrometry-based proteomics including AP-MS on endogenously tagged cell lines and SEC-MS for proteome-wide complex mapping. PPI networks processed using node2vec deep learning model to reduce dimensionality, producing PPI embeddings containing information about protein interactions.
cm4ai:acquisition:2
Single-Cell RNA Sequencing for Perturbation Mapping: Single-cell RNA sequencing using 10x Genomics 3'HT kit to capture transcriptional states following CRISPR perturbations. Generates genome-scale perturbation cell atlas mapping transcriptional and fitness phenotypes. Raw sequence data deposited to NCBI BioProject/Sequence Read Archive (SRA).
cm4ai:acquisition:3
MuSIC Pipeline Integration: Multi-Scale Integrated Cell (MuSIC) pipeline integrates PPI embeddings and image embeddings using contrastive deep learning to obtain co-embeddings for each protein. Community detection performed based on all-by-all similarities of protein pairs in co-embedding space, producing hierarchical cell maps as final output. Maps annotated using Gene Ontology, Reactome pathways, and large language model approaches.
cm4ai:acquisition:4
DescriptionIDPreprocessing Details
Deep Learning Embedding Generation: PPI networks processed using node2vec deep learning model to reduce dimensionality and produce PPI embeddings. IF images processed using Human Protein Atlas deep learning model to reduce dimensionality and produce image embeddings. PPI and image embeddings integrated to obtain co-embeddings using contrastive deep learning, learning co-embeddings such that original embeddings can be reconstructed with minimal information loss.
cm4ai:preproc:1node2vec deep learning for PPI network dimensionality reduction, Human Protein Atlas deep learning model for image embedding, Contrastive deep learning for PPI and image embedding integration, Co-embedding optimization to minimize information loss
Hierarchical Community Detection: Community detection performed on co-embedding space using multiscale community detection algorithms implemented in Cytoscape. Produces hierarchical directed acyclic graphs (DAG) of protein assemblies at multiple resolutions, with 10 layers of depth representing communities from large cell compartments to small protein complexes.
cm4ai:preproc:2All-by-all similarity computation in co-embedding space, Multiscale community detection using Cytoscape algorithms, Hierarchical structure generation as directed acyclic graphs, 10-layer depth hierarchy from compartments to complexes
Cell Map Annotation: Two-pronged annotation approach: (1) Alignment to known protein function and pathway resources including Gene Ontology (GO) and Reactome to determine protein assemblies with high overlap with known cell biology, and (2) Large language model (LLM) approach to name sets of proteins and assign name confidence scores.
cm4ai:preproc:3Alignment to Gene Ontology for functional annotation, Alignment to Reactome pathways for pathway annotation, LLM-based naming of protein assemblies with confidence scores, Validation of assemblies against known cell biology
Quality Control for Imaging Data: Standardized automated fixation and permeabilization protocols using pipetting robot. Consistent staining protocols across conditions using Human Protein Atlas antibodies. Quality control of imaging data before release and processing through MuSIC pipeline.
cm4ai:preproc:4Automated protocols for consistency, Standardized staining across all conditions, Quality checks before downstream processing, Release only after quality validation
Quality Control for Mass Spectrometry Data: Quality control and validation of AP-MS and SEC-MS data before public release. Mass spectrometry data for human iPSCs deposited to MassIVE Repository, and data for human cancer cells also deposited to MassIVE Repository.
cm4ai:preproc:5QC procedures for AP-MS data, QC procedures for SEC-MS data, Validation before public release, Deposition to MassIVE repositories with persistent identifiers
Integrative Structure Modeling: Bioinformatics pipeline for annotating MuSIC communities by available structural information about community members and their interactions. Structural information includes PDB, AlphaFold Protein Structure Database, crosslinking mass spectrometry, and prediction of disordered sequence segments. Communities ranked by structural information amount as proxy for integrative modeling feasibility. Modeling protocol scripted using Python Modeling Interface package based on Integrative Modeling Platform (IMP) version 2.18.
cm4ai:preproc:6PDB structural information integration, AlphaFoldDB structure integration, Crosslinking mass spectrometry data incorporation, Disordered region prediction using sequence analysis, Feasibility ranking for structure modeling, IMP-based structural model generation
Cleaning DetailsDescriptionID
RO-Crate packaging with metadata and provenance, JSON-Schema validation of all datasets, ARK persistent identifier assignment, Provenance graph computation and linking, Machine-readable metadata in JSON-LD with Schema.org, EVI vocabularies
FAIRSCAPE AI-Readiness Packaging: All datasets packaged using FAIRSCAPE framework which creates RO-Crate packages with datasets, metadata, provenance graphs, and software. FAIRSCAPE-CLI validates inputs and creates output RO-Crate packages. FAIRSCAPE server assigns persistent resolvable globally unique identifiers (ARK scheme), decomposes RO-Crates into components, and computes end-to-end provenance entailments using EVI Evidence Graph Ontology.
cm4ai:cleaning:1
GO mapping for protein functions, Reactome mapping for pathways, PDB and AlphaFold for structural information, Schema.org and EVI for metadata, Controlled vocabulary mapping for all keywords, NCI Thesaurus, BAO, Cell Ontology, CHEBI ontology mappings
Data Standard Mapping: Data mapped to applicable standards and ontologies including Gene Ontology (GO), Reactome, Protein Data Bank (PDB), AlphaFold Protein Structure Database, Schema.org, and EVI Evidence Graph Ontology. Keywords mapped to controlled vocabularies from NCI Thesaurus, BioAssay Ontology, Cell Ontology, CHEBI, EFO, and other ontologies.
cm4ai:cleaning:2
  • ID
    cm4ai:maintainer:1
    Description
    CM4AI Consortium: Multidisciplinary consortium managing dataset maintenance including University of California San Diego (lead), University of California San Francisco, Stanford University, University of Virginia, Yale University, University of Alabama at Birmingham, Simon Fraser University, and The Hastings Center. Data Governance Committee led by Jillian Parker (jillianparker@health.ucsd.edu). Ethical Review by Vardit Ravitsky (ravitskyv@thehastingscenter.org) and Jean-Christophe Belisle-Pipon (jean-christophe_belisle-pipon@sfu.ca).
    Maintainer Details
    • University of California San Diego (lead institution, Ideker Lab)
    • UCSF (Krogan Lab - protein interactions, Sali Lab - structure modeling)
    • Stanford University (Lundberg Lab - spatial proteomics)
    • University of Virginia (Clark Lab - standards and FAIRSCAPE)
    • Yale University (Schulz Lab - workforce development)
    • University of Alabama at Birmingham (Chen Lab - teaming, U-BRITE platform)
    • Simon Fraser University (Bélisle-Pipon - ethics)
    • The Hastings Center (Ravitsky - ethics)
    • Data Governance Committee (Jillian Parker, jillianparker@health.ucsd.edu)
ID
cm4ai:retention:1
Description
Digital data maintained according to NIH data sharing policies with long-term preservation in University of Virginia's LibraData repository supported by committed institutional funds. No planned sunset for data availability. Archived RO-Crates with persistent identifiers (ARK, future DOIs) ensure long-term accessibility and citability.
Retention Details
  • NIH data sharing policy compliance
  • UVA Dataverse institutional commitment
  • Persistent identifiers (ARK, future DOIs)
  • No planned data sunset
  • Machine-readable metadata for long-term discoverability
  1. ID
    cm4ai:sensitive:1
    Description
    Cell Line Origin Metadata: While cell lines are de-identified and cannot be matched to specific individuals, metadata about cell line origins (age, sex, race of original donor) is retained for scientific context. This metadata does not constitute identifiable human subjects data under current knowledge. Data derived from commercially available de-identified human cell lines and does not represent all biological variants in the population at large.
    Sensitive Elements Present
    False
    Sensitivity Details
    • De-identified commercial cell lines
    • Donor demographic metadata for scientific context only
    • Cannot be matched to individuals with current knowledge
    • Ethically sourced from ATCC and HipSci
    • Does not represent all biological variants seen in the population
DescriptionExternal ResourcesID
CM4AI Project Website: Official project website and data portal using U-BRITE platformhttps://www.cm4ai.orgcm4ai:resource:1
NIH RePORTER Project Details: Federal grant information and project details for Bridge2AI Functional Genomicshttps://reporter.nih.gov/project-details/11211616cm4ai:resource:2
University of Virginia Dataverse: LibraData repository with archived RO-Crates and data releaseshttps://doi.org/10.18130/V3/DXWOS5, https://doi.org/10.18130/V3/B35XWX, https://doi.org/10.18130/V3/F3TD5R, https://doi.org/10.18130/V3/K7TGEMcm4ai:resource:3
Nature Publication: Schaffer LV, Hu M, Qian G, et al. Multimodal cell maps as a foundation for structural and functional genomics. Nature. Published April 9, 2025. https://doi.org/10.1038/s41586-025-08878-3cm4ai:resource:4
bioRxiv Preprint: Clark T, et al. Cell Maps for Artificial Intelligence: AI-Ready Maps of Human Cell Architecture from Disease-Relevant Cell Lines. BioRXiv, May 2024. https://doi.org/10.1101/2024.05.21.589311cm4ai:resource:5
FAIRSCAPE Framework Documentation: AI-readiness framework documentation, tutorial, and installation instructionshttps://fairscape.github.iocm4ai:resource:6
Integrative Modeling Platform: Open source IMP package for integrative structure modelinghttp://integrativemodeling.orgcm4ai:resource:7
Network Data Exchange (NDEx): Repository and visualization platform for cell maps and networkshttps://www.ndexbio.orgcm4ai:resource:8
Bridge2AI Program: Parent NIH Common Fund program supporting AI-ready biomedical datasetshttps://commonfund.nih.gov/bridge2aicm4ai:resource:9
NIH Common Fund Data Ecosystem (CFDE): Collaboration partner for data curation and integrationhttps://www.nih-cfde.orgcm4ai:resource:10
MassIVE Proteomics Repository: Mass spectrometry data repository for iPSC and cancer cell SEC-MS dataMassIVE Repository (SEC-MS human iPSCs), MassIVE Repository (SEC-MS human cancer cells)cm4ai:resource:11
NCBI Sequence Read Archive: Repository for CRISPR perturbation screen raw sequence dataNCBI BioProject, Sequence Read Archive (SRA)cm4ai:resource:12
Perturbation Cell Atlas Publication: Nourreddine S, Doctor Y, Dailamy A, et al. A PERTURBATION CELL ATLAS OF HUMAN INDUCED PLURIPOTENT STEM CELLS. bioRxiv. 2024 Nov 4. PMCID: PMC11580897 https://doi.org/10.1101/2024.11.03.621734cm4ai:resource:13
🚀

Uses

What (other) tasks could the dataset be used for?

DescriptionID
Integrate multimodal data streams (spatial proteomics via IF imaging, protein-protein interactions via AP-MS and SEC-MS, and genetic perturbations via CRISPR screens) using the Multi-Scale Integrated Cell (MuSIC) software pipeline employing deep learning models and community detection algorithms to produce hierarchical cell maps.
cm4ai:task:1
Enable development of visible neural networks (VNNs) and visible machine learning tools that use hierarchical cell maps as interpretable structures for AI model architectures, allowing interrogation of how protein assemblies affect cell-level phenotypes and interpretation of genetic variants and mutations.
cm4ai:task:2
Characterize cell architecture and protein interactions in disease-relevant cell lines including treated and untreated MDA-MB-468 breast cancer cells (with paclitaxel and vorinostat) and differentiated and naive KOLF2.1J induced pluripotent stem cells (iPSCs) differentiated into neurons and cardiomyocytes.
cm4ai:task:3
Develop ethical AI frameworks and governance structures for biomedical data, including Value-Sensitive Design methodologies, axiological repositories, CM4AI Life Cycle framework, and guidelines for responsible design of datasets and AI technologies.
cm4ai:task:4
Perform integrative structure modeling of MuSIC protein communities to determine structural models using data from PDB, AlphaFoldDB, crosslinking mass spectrometry, and prediction of disordered sequence segments, enabling structural and functional genomics applications.
cm4ai:task:5
DescriptionID
AI Model Training for Functional Genomics: Primary intended use is training and development of artificial intelligence and machine learning models for functional genomics research. AI-ready datasets with full provenance, metadata, and validation enable immediate use in AI/ML pipelines without reformatting.
cm4ai:use:1
Visible Neural Network Development: Development of visible neural networks (VNNs) that use hierarchical cell maps as interpretable model architectures. Unlike black box models, VNNs built on cell maps allow interrogation of how protein assemblies affect cell-level phenotypes, enabling interpretation of genetic variants and mutations in the context of cellular mechanisms.
cm4ai:use:2
Genotype-Phenotype Mapping Research: Research into interpretable genotype-phenotype learning using multi-scale cell maps. Enables understanding of mechanisms by which genotypes translate to phenotypes, supporting precision medicine applications and genomic variant interpretation.
cm4ai:use:3
Drug Response and Synergy Prediction: Analysis of cellular responses to drug treatments (paclitaxel, vorinostat) to predict drug response and synergy. Cell maps under different treatment conditions enable visible machine learning for drug discovery and personalized medicine applications.
cm4ai:use:4
Disease Mechanism Research: Study of disease mechanisms in cancer, neuropsychiatric disorders, and cardiac disorders through analysis of chromatin modifiers and metabolic enzymes in disease-relevant cell contexts. Supports understanding of disease pathways and identification of therapeutic targets.
cm4ai:use:5
Structural and Functional Genomics: Use as foundation for integrative structure modeling of protein communities, combining multimodal cell maps with PDB, AlphaFoldDB, and crosslinking mass spectrometry data to determine structural models of protein assemblies, as demonstrated in Nature publication (Schaffer, Hu et al., April 2025).
cm4ai:use:6
Model for AI-Ready Biomedical Dataset Development: Use as exemplar for future AI-ready biomedical dataset development, demonstrating best practices in FAIR principles implementation, provenance tracking, ethical data governance, and AI-readiness packaging using RO-Crate and FAIRSCAPE frameworks.
cm4ai:use:7
DescriptionID
Clinical Decision-Making Without Validation: Laboratory data from cell lines are not to be used in clinical decision-making or any context involving patient care without appropriate regulatory oversight and approval. Requires domain expertise and clinical validation before any clinical applications. Explicitly prohibited per dataset terms.
cm4ai:discouraged:1
Use During Incomplete Data Release: This is an interim/beta release with data not yet in completed final form. Some datasets are under temporary pre-publication embargo, protein interrogation sets incompletely overlap across data modalities, and computed cell maps not yet included in releases. Full integration and final cell maps will be available in future releases through November 2026.
cm4ai:discouraged:2
Analysis Without Domain Expertise: Datasets require domain expertise for meaningful analysis and interpretation. Current release is most suitable for bioinformatics analysis of individual datasets. Not suitable for use without understanding of functional genomics, proteomics, cell biology, and AI/ML methodologies. Training resources available through CM4AI Skills and Workforce Development module.
cm4ai:discouraged:3
ID
cm4ai:license:1
Description
Data licensed for reuse under Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International license (https://creativecommons.org/licenses/by-nc-sa/4.0/). Attribution is required to the copyright holders and the Cell Maps for Artificial Intelligence project. Any publications referencing this data or derived products should cite the Nature article (Schaffer LV, Hu M, et al. Multimodal cell maps as a foundation for structural and functional genomics. Nature. 2025. doi:10.1038/s41586-025-08878-3) and the bioRxiv preprint (Clark T, et al. Cell Maps for Artificial Intelligence: AI-Ready Maps of Human Cell Architecture from Disease-Relevant Cell Lines. BioRXiv, May 2024. doi:10.1101/2024.05.21.589311) and directly cite the data collection. Commercial use requires separate license negotiation with copyright holder (UCSD, Stanford, and/or UCSF depending upon specific data package). A Data Access Committee (led by Jillian Parker) supervises ethical matters related to dataset distribution and potential dual licensing for commercial use. Copyright (c) 2025 The Regents of the University of California except where otherwise noted. Spatial proteomics raw image data is copyright (c) 2025 The Board of Trustees of the Leland Stanford Junior University.
License Terms
  • Attribution required to copyright holders and authors
  • Non-commercial use only (commercial requires separate license)
  • Share-alike - derivative works must use same license
  • Must cite Nature publication and bioRxiv preprint and data collection DOI
  • Data Access Committee oversight for ethical distribution
📤

Distribution

How will the dataset be distributed?

CC BY-NC-SA 4.0
🔄

Maintenance

How will the dataset be maintained?

ID
cm4ai:updates:1
Description
Dataset regularly updated and augmented through end of project in November 2026. Beta releases on quarterly basis with periodic data augmentation. Initial alpha release (v0.5) provided as supplemental data. March 2025 Beta (V1.4) includes perturb-seq in KOLF2.1J iPSCs, SEC-MS in iPSCs and derivatives, and IF images in MDA-MB-468 under three conditions. June 2025 Beta (V2.1) revision adds RGB IF images, ro-crate metadata corrections, and naming convention changes, plus SEC-MS for MDA-MB-468. October 2025 Beta adds Perturb-seq for MDA-MB-468 breast cancer cells and additional SEC-MS data. Future releases will include computed cell maps and complete integration of all data streams. Long-term preservation in University of Virginia Dataverse with committed institutional support.
Frequency
Quarterly updates through November 2026; long-term preservation thereafter
Update Details
  • Alpha release v0.5 (supplemental data)
  • March 2025 Beta release V1.4 (doi:10.18130/V3/B35XWX)
  • June 2025 Beta release V2.1 (doi:10.18130/V3/F3TD5R)
  • October 2025 Beta release (doi:10.18130/V3/K7TGEM)
  • Quarterly augmentation through November 2026
  • Future releases to include computed cell maps
  • Final release expected November 2026
  • Long-term preservation in UVA Dataverse
👥

Human Subjects

Does the dataset relate to people?

ID
cm4ai:hsr:1
Description
CM4AI data are distinctive within Bridge2AI in that they are non-clinical data from tissue cultures and are considered to be de-identified as they cannot be matched, with current knowledge, to a human subject. Both cell lines (MDA-MB-468 and KOLF2.1J) are commercially available, ethically sourced, de-identified cell lines. MDA-MB-468 available from ATCC. KOLF2.1J available from HipSci resource for non-profit organizations via simple MTA. Human Subjects: No. De-identified Samples: Yes. FDA Regulated: No.
Involves Human Subjects
False
IRB Approval
  • Not applicable - de-identified cell lines from commercial sources
Ethics Review Board
  • CM4AI Ethics Module (Vardit Ravitsky, Jean-Christophe Bélisle-Pipon)
  • Data Access Committee (Jillian Parker)
  • Bridge2AI Ethics Working Group participation
Special Populations
  • MDA-MB-468 derived from 51-year-old black female (de-identified)
  • KOLF2.1J derived from healthy male Northern European donor (de-identified)
Generated on 2026-04-15 17:53:10 using Bridge2AI Data Sheets Schema