Hugging Face's logo

Datasets:
DataTecnica
/
RoP_biomedical

Tasks:
Feature Extraction
Modalities:
Tabular
Text
Formats:
parquet
Languages:
English
Size:
1M - 10M
Tags:
biomedical
harmonization
common-data-elements
ontology
FAIR
interoperability
DOI:
Libraries:
Datasets
pandas
Polars
License:
Dataset card Data Studio Files
xet
 Community
1
RoP_biomedical / README.md
MikeDT's picture
MikeDT
Upload README.md with huggingface_hub
c53e49f verified
preview code
|
raw
history blame contribute delete
21.2 kB
metadata 
license: other
license_name: cc-by-nc-4.0-data-agplv3-code
license_link: https://creativecommons.org/licenses/by-nc/4.0/
task_categories:
  - feature-extraction
language:
  - en
tags:
  - biomedical
  - harmonization
  - common-data-elements
  - ontology
  - FAIR
  - interoperability
  - AI-readiness
  - genomics
  - clinical-data
  - neurodegeneration
  - OMOP
  - LOINC
  - ICD-10
size_categories:
  - 1M<n<10M
pretty_name: RoP - Biomedical Reference of Parameters v2026.04
configs:
  - config_name: default
    data_files:
      - split: train
        path: v2026.04/elements.parquet

RoP — Biomedical Reference of Parameters

One CDE set to rule them all...

License: CC-BY-NC-4.0 License: AGPL-3.0 DOI

📋 Interested in help using RoP? → Submit Interest Form ← Start here

Executive Summary

RoP decreases activation energy in biomedical research by accelerating progress past the biggest bottlenecks ... data wrangling, interoperability and AI-readiness.

Right now, multi-cohort studies spend months (often years) manually mapping variables—PPMI calls it "MoCA Total Score," NACC calls it MOCATOTS, ADNI calls it MOCA. Same assessment, zero automatic compatibility. Multiply this by 500 variables across 10 cohorts, and you're looking at person-years of manual mapping before analysis even starts.

RoP provides 1.33 million pre-harmonized Common Data Elements (CDEs) organized into 13 themes covering every layer of biomedical data—from individual identity and clinical phenotypes to genomics, imaging, governance, and resource catalogs. Built on OMOP, LOINC, ICD-10, HPO, Mondo, and 9 other major vocabularies. Tag your variable with a RoP ID once—instant compatibility with 200+ studies.

This isn't aspirational. It's running in production across hundreds of thousands of samples and multiple millions of data points for collaborators leading massive federated open science initiatives.

The work behind RoP: Weeks of planning, one weekend of compute, decades of experience dealing with biomedical data harmonization challenges. The v2026.04 release represents 6+ person-years of expert harmonization work, distilled into a 7.8 GB download.

What makes this different:

  • Living collection — quarterly releases with new vocabularies, collections, and improvements (not static PDFs)
  • AI-powered — semantic embeddings enable instant CDE matching in The Forge platform (<10 minutes vs 6 months manual)
  • FAIR by design — Findable (DOI-versioned), Accessible (open download), Interoperable (cross-vocab mappings), Reusable (full build reproducibility)
  • Community-governed — human-in-the-loop validation ensures quality at scale

Technical depth: 1.39 million common data model components harmonized into 1.33 million interoperable data elements, mapped to biomedical language model embeddings representing over 4,000 semantic clusters. Likely all the data points you'll ever need for interoperability and AI-readiness in biomedical research, with a particular focus on neurodegeneration.

RoP v2026.04 Summary

What Makes RoP Different

Most CDE projects are static PDFs. A committee meets, agrees on definitions, publishes a table, and hopes the field adopts it. Updates take years. Interoperability is aspirational.

RoP is an AI-powered, FAIR-compliant, governed data infrastructure. Not another CDE table — a living semantic ecosystem that:

🔗 Integrates 1.33 Million CDEs Across 9 Standards

RoP unifies definitions from every major biomedical vocabulary — OMOP, LOINC, ICD-10, RxNorm, HPO, Mondo, NINDS-CDE, PhenX, CDISC, DICOM, BIDS, DUO — into a single harmonized reference. No more choosing between standards: RoP preserves cross-vocabulary mappings as first-class data.

Example: Looking for "Parkinson's disease"? RoP connects:

  • ICD-10-CM: G20 (Parkinson's disease)
  • OMOP: 378419 (Primary parkinsonism)
  • HPO: HP:0001300 (Parkinsonism)
  • Mondo: MONDO:0005180 (Parkinson disease)
  • NINDS-CDE: ParkinsonsDisease_DiagnosticStatus

One variable tag in RoP gets you interoperability with all five vocabularies.

🤖 AI-Powered Semantic Matching

RoP ships with SapBERT embeddings for all 1.33M CDEs — biomedical language model vectors trained on PubMed abstracts. The Forge uses these embeddings to:

  1. Auto-suggest matches: Upload "MoCA Total Score" → Forge instantly finds MoCA_TotalScore and 47 equivalent CDEs across NACC, NINDS-CDE, and PhenX
  2. Catch errors: Flag likely unit mismatches, implausible value ranges, consent code violations
  3. Learn from curation: Human-in-the-loop corrections improve the model for everyone

No manual crosswalks. No months of expert panels. Just upload your data dictionary and let AI + human review handle the rest.

🛡️ Governed at Ingest Time via The Forge

RoP integrates with DataTecnica's The Forge — a governance platform that enforces RoP conformance when data enters your system (not "after the paper is published").

Forge workflow:

  1. Data team uploads cohort file
  2. Forge matches variables to RoP IDs using semantic embeddings
  3. Curator reviews AI suggestions, approves/corrects mappings
  4. Forge validates: consent codes (DUO), value ranges, required fields
  5. Non-compliant data gets blocked at ingest with specific error messages
  6. Approved data flows into your analysis pipeline, pre-harmonized

This is what "governance" actually means: Automated quality checks + human oversight + enforced standards.

📊 FAIR by Design

RoP bundles are Findable, Accessible, Interoperable, Reusable out of the box:

  • Findable: DOI-versioned releases on Zenodo + Hugging Face
  • Accessible: Open download (CC-BY-NC-4.0), no paywalls, no restricted vocabularies
  • Interoperable: Parquet format, FAISS index, semantic embeddings, cross-vocab xrefs
  • Reusable: AGPLv3 code, full build scripts, SHA256 checksums, reproducible pipelines

FAIR isn't a checklist — it's the architecture. Every RoP release includes machine-readable provenance (manifest.json), cryptographic integrity checks, and full build reproducibility scripts.

🔄 Continuously Growing, Versioned Quarterly

Traditional CDEs: Committee publishes v1.0 in 2020 → Errata in 2022 → Maybe v2.0 in 2025

RoP: Quarterly releases (v2026.04, v2026.07, v2026.10, v2027.01, ...) with:

  • New vocabularies (e.g., SNOMED-free alternatives, emerging ontologies)
  • New collections (GP2, CARD, Answer ALS → your project next)
  • Improved embeddings (BioLORD-2023, domain-tuned models)
  • Bug fixes, schema migrations, governance updates

Pin the version in your methods section (RoP v2026.04) for reproducibility. Upgrade when you're ready.

🧬 Multi-Modal & Multi-Scale

RoP covers every layer of biomedical data in 13 themes:

What Examples RoP Theme
Clinical phenotypes Diagnoses, symptoms, drugs, lab tests 12 (Clinical Concepts)
Genomics Assay types, platforms, pipelines, variants 6 (Omics)
Imaging DICOM tags, BIDS metadata, scanner parameters 7 (Imaging)
Biosample Tissue types, cell lines, anatomical sites 5 (Biosample)
Assessments MoCA, UPDRS, MMSE, CDR instruments 8 (Clinical Instruments)
Governance Consent codes, data use, sharing controls 9 (Governance)
Meta-analysis GWAS summary stats, PCA loadings 11 (Summary Statistics)
Discoverability Study catalogs, resource metadata 13a (Resources)

Cross-modal integration is built-in: Link imaging scans (Theme 7) → genomic assays (Theme 6) → clinical diagnoses (Theme 12) → biosample provenance (Theme 5) using common individual IDs (Theme 1).

Beyond biomedicine: The harmonization pattern (AI-assisted HitL + federated governance + versioned schemas) transfers cleanly to other verticals. Applications underway with collaborators in stock trading (social data + market integration) and motorsports (safety, performance, marketing integration).

The Problem RoP Solves

You spend 6 months harmonizing data that should have been compatible from day one.

Every biomedical consortium faces this:

  • PPMI calls it "MoCA Total Score"
  • NACC calls it MOCATOTS
  • ADNI calls it MOCA
  • Your site calls it moca_total

Same assessment, four different variable names, no automatic way to merge them. Multiply this by 500 variables across 10 cohorts, and you're looking at person-years of manual mapping before analysis even starts.

RoP fixes this: Tag your moca_total with RoP ID MoCA_TotalScore once. Instant compatibility with PPMI, NACC, ADNI, and 200+ other studies using the same RoP tag.

Who Made This & Why

Primary Authors: Pietro Marini, Alan Long, Hirotaka Iwaki, Mike Nalls, Dan Vitale

Organization: DataTecnica

Collaborators: Mette Peters, Hampton Leonard, Andy Henrie, Amara Alexander, Elise Marsan, Yang Fann, Mark Cookson, Cornelis Blauwendraat, Andy Singleton, Huw Morris, Tim Hohman, Sara Biber, John Crary, Syed Shah, Brittany Dugger, David Gutman, Chris Morris, Pat Brannelly, Liesel Jones, Mat Koretsky, Cole Tindall, Mukta Phatak, Zane Jaunmuktane, Mimi Tambi, Brandon Jernigan, Terri Thompson, Mike Karlovich, Kurt Farrell, and many more... CDEs are a community effort.

The "Why":

We want to accelerate the biomedical research community. Let's make research as FAIR as possible.

We built RoP to solve this for ourselves. We're releasing it open source for non-commercial use (AGPLv3 + CC-BY-NC-4.0) because the biomedical community needs a shared CDE reference that:

  1. Works today (not after the next committee meeting)
  2. Enforces adoption (via Forge validation at ingest)
  3. Grows continuously (quarterly releases, not 5-year cycles)
  4. Distributes freely (no SNOMED/MedDRA/CPT fees)
  5. Leverages AI (semantic embeddings, not keyword search)
  6. Implements FAIR (as infrastructure, not aspiration)

Quick Start

I'm a Biomedical Researcher (Not a Programmer)

You don't need to code to use RoP. Here's how to get started:

  1. Submit Interest Form --> Let's figure out how we can help you with your project(s)
  2. We'll help you map your data dictionary to RoP CDEs using The Forge platform (AI-assisted, human-validated)
  3. Your harmonization happens in <10 minutes (vs 6 months manual crosswalk)

Found an error or need a new CDE? Use the Schema Feedback Form to report corrections or suggest additions.

No downloads, no command line, no setup required. The Forge handles everything.

Example workflow:

  • Upload your data dictionary (Excel/CSV)
  • AI suggests matches: "Your moca_total → RoP's MoCA_TotalScore (98% confidence)"
  • Expert reviews AI suggestions (10 minutes)
  • Download harmonized mapping → apply to your cohort
  • Done. Now interoperable with 200+ studies.

I'm a Data Scientist or Developer

Download Pre-built Bundle (Recommended)

Hugging Face (fastest, CDN-backed):

pip install huggingface_hub
python3 -c "
from huggingface_hub import hf_hub_download
for f in ['elements.parquet', 'embeddings.npy', 'embeddings.faiss', 'manifest.json']:
    hf_hub_download('DataTecnica/RoP_biomedical', f'v2026.04/{f}', cache_dir='./rop_bundle')
"

Zenodo (DOI-versioned, archival):

wget https://huggingface.co/datasets/DataTecnica/RoP_biomedical/resolve/main/v2026.04/
tar -xzf rop_v2026.04.tar.gz
cd rop_v2026.04

Build from Source (Reproducible)

Requires: Python 3.10+, 32 GB RAM, CUDA GPU (optional, speeds up embedding from 30 hours to 6 hours)

git clone https://github.com/datatecnica/rop_build.git
cd rop_build
python3 -m venv venv && source venv/bin/activate
pip install -e .

# Download 9 public sources (~6 GB)
python3 scripts/sprint1_download_all.py

# Build full bundle
python3 scripts/sprint1_dedup_pass1.py           # Dedup: 2 min
python3 scripts/generate_embeddings_direct.py    # Embed: 6 hrs (GPU) / 30 hrs (CPU)
python3 scripts/build_faiss_index.py             # Index: 25 min
python3 scripts/package_final_bundle.py          # Package with boutique CDEs: 1 min

# Output: dist/rop_v2026.04/ (7.8 GB)

What's Inside: The Bundle 

v2026.04/
├── elements.parquet         1,328,973 CDEs (151 MB)
├── embeddings.npy           SapBERT 768-dim vectors (3.9 GB)
├── embeddings.faiss         IVF4096 similarity index (3.9 GB)
└── manifest.json            SHA256 checksums + metadata

Verify integrity after download:

sha256sum -c <(jq -r '.files | to_entries[] | "\(.value.sha256)  \(.key)"' manifest.json)

Real-World Use Cases

1. Harmonize Multi-Cohort Studies

Map your 500 variables to RoP IDs → merge data from PPMI, ADNI, NACC, Answer ALS with zero manual crosswalks.

2. Build a FAIR Data Catalog

Use Theme 13a (Resources) to tag studies with standardized metadata: sample size, modalities, consent codes, FAIR scores.

3. Enforce Consent & Governance

Use Theme 9 + Forge to block data uploads that violate DUO consent codes or sharing restrictions — before data enters your system.

4. Run Meta-Analyses Faster

Theme 11 (Summary Stats) harmonizes GWAS covariates across cohorts: no more "which ancestry PCs did they use?" guesswork.

5. Link Multi-Modal Data

Connect genomics (Theme 6) + imaging (Theme 7) + clinical phenotypes (Theme 12) with common individual IDs (Theme 1).

Integration with The Forge

RoP + Forge = Governed, AI-Assisted Data Pipelines

  • RoP = The reference standard (1.33M harmonized CDEs)
  • The Forge = The enforcement layer (validates, matches, curates)

Forge uses RoP to:

  1. Match your variables to RoP IDs using semantic embeddings (AI)
  2. Vet matches via human-in-the-loop curation (governance)
  3. Validate consent codes, value ranges, required fields (FAIR)
  4. Block non-compliant data at ingest (enforcement)
  5. Grow RoP with approved project collections (continuous improvement)

Submit Interest Form to discuss Forge integration for your project.

The 13 Themes (224 Anchor CDEs)

Theme Coverage Anchors Example CDEs
1. Identity Individual IDs, aliases, external linkages 3 IndividualID, IndividualAliases
2. Time Visit dates, age, years from diagnosis 7 VisitDate, AgeAtVisit, YearsFromEnrollment
3. Sex Chromosomal, assigned-at-birth, gender identity 3 ChromosomalSex, GenderIdentity
4. Ancestry & Pedigree Race/ethnicity, genetic ancestry, family structure 18 RaceOMB, GeneticAncestry, FamilyID
5. Biosample Sample types, anatomy, cell types 20 BiosampleID, TissueType, AnatomicalSite
6. Omics Assay methods, platforms, pipelines 22 LOINCComponent, SequencingPlatform
7. Imaging DICOM/BIDS metadata, scanner parameters 14 ImagingModality, FieldStrength, DICOMSeriesUID
8. Clinical Instruments MoCA, UPDRS, MMSE, CDR 13 MoCA_TotalScore, UPDRS_PartIII
9. Governance Consent codes, sharing controls 27 DUOConsentCode, ReshareAllowed
10. Data Assets File URIs, formats, checksums 15 DataAssetURI, FileFormat, SHA256Checksum
11. Summary Stats GWAS, PCA, meta-analysis metadata 27 GWAS_PValue, PCA_PC1Loading
12. Clinical Concepts Diagnoses, drugs, labs, procedures 13 ClinicalConceptCode (OMOP backbone)
13a. Resources Study catalogs, repository metadata 26 StudySampleSize, ResourceFAIRScore
13b. Collections Project-specific CDE bundles 16 CollectionID, MemberCDECount

Total: 1,328,973 elements in v2026.04 (224 anchors + 1.33M harmonized CDEs from 9 sources + 9 boutique collections)

Vocabulary Inclusion Policy: Open Distribution Only

RoP includes only vocabularies that allow free redistribution (CC-BY, CC0, Apache-2.0, public domain).

✅ Included (open): LOINC, HPO, Mondo, DUO, ICD-10/11, RxNorm, ATC, MeSH, UCUM, CDISC, NINDS-CDE, PhenX, BIDS, DICOM, ClinVar, HGNC, NCIt, OncoTree, HemOnc, NAACCR, CVX, NDFRT, and 40+ more via OMOP/Athena.

❌ Excluded (commercial licenses): SNOMED CT, MedDRA, CPT4, UMLS CUIs — these require vendor fees for redistribution. RoP can cross-reference their IDs when upstream sources (NINDS-CDE, OMOP) provide mappings, but no commercial vocabulary content ships in RoP bundles.

This prioritizes redistributability over maximal coverage. Users with SNOMED/MedDRA licenses can layer them onto RoP in their own deployments.

Documentation

Resource Description
docs/SPEC.md Full interoperability contract specification
docs/ANCHORS.md 224 anchor CDE reference with rationale
docs/GOVERNANCE.md Forge governance model + contribution workflow
docs/DISTRIBUTION.md GitHub + Hugging Face + Zenodo strategy
docs/ROADMAP.md Quarterly release schedule, planned features

Citation 

@dataset{rop_v202604,
  author       = {Marini, Pietro and
                  Long, Alan and
                  Iwaki, Hirotaka and
                  Nalls, Michael A. and
                  Vitale, Dan},
  title        = {RoP v2026.04 - Biomedical Reference of Parameters},
  year         = {2026},
  publisher    = {Hugging Face},
  doi          = {10.57967/hf/8781},
  url          = {https://huggingface.co/datasets/DataTecnica/RoP_biomedical}
}

License

  • Data (CDEs, embeddings): CC-BY-NC-4.0 — Free for non-commercial use with attribution
  • Code (Python, scripts): AGPLv3 — Open source, copyleft, non-commercial

Commercial Use: Requires separate licensing. Contact info@datatecnica.com for commercial terms.

Attribution:

This work uses RoP (Biomedical Reference of Parameters) v2026.04, © 2026 Dan Vitale, Pietro Marini, Michael A. Nalls (DataTecnica), licensed under CC-BY-NC-4.0 (data) + AGPLv3 (code). https://doi.org/10.57967/hf/8781

Community & Support

Governance: All proposals (new anchors, sources, vocabularies) reviewed via Forge's human-in-the-loop workflow. See docs/GOVERNANCE.md.

Acknowledgments

RoP builds on work from:

  • NIH CARD (Center for Alzheimer's and Related Dementias)
  • GP2 (Global Parkinson's Genetics Program)
  • NACC (National Alzheimer's Coordinating Center)
  • Answer ALS, SEA-AD, ADSP-PHC, ASAP, BDR, BDSA, PART through their connection with the Path-ND Consortium by the 10,000 Brains Project

Upstream standards: OHDSI (OMOP), Regenstrief (LOINC), Monarch Initiative (HPO, Mondo), NIH (NINDS-CDE, PhenX), CDISC, GA4GH (DUO), DICOM, BIDS.

Special recognition: We are deeply grateful to the National Library of Medicine (NLM) for their decades of foundational work in biomedical informatics and vocabulary development. NLM's efforts (including UMLS, MeSH, RxNorm, and countless other standards) have been instrumental in making semantic interoperability possible. We would be honored if any of our harmonization work could contribute back to NLM's mission in any way.

Foundational concepts for this work are based on this preprint (Long et al 2024, https://pubmed.ncbi.nlm.nih.gov/39484274/) hopefully in press very soon.

Built with 🔨 by DataTecnica | datatecnica.com