Come for the dataset. Leave with the science.
Search the catalog, check a card in 30 seconds and request access with your analysis plan. Your pipeline runs on your own compute, reading the dataset in place, and approved outputs land in your storage. LooporaData never receives a copy.
For research teams
How it works, step by step
- Search the catalog: Filter by modality, disease area, access route and permitted use. Open datasets and rare cohorts sit in one catalog.
- Check a card in 30 seconds: People, not samples. Commercial use, model training and whether weights may leave read yes, no or unknown, and every field is labelled verified, holder stated or unknown.
- Request access with your plan: Send your analysis plan and output manifest: the files your pipeline will write. The data holder approves the request and the manifest.
- Run on your own compute: Your pipeline runs in an attested container on your own compute: Docker anywhere, or Apptainer on HPC. It reads the dataset in place with short-lived credentials scoped to your project.
- Take out approved outputs: Approved outputs land in your storage, with their context: datasheet version, consent codes, pipeline fingerprint, output manifest and audit record.
What you take out
The dataset stays with the data holder. You leave with the science: the statistics, associations, signatures and evaluation results you need to publish, file or build on.
- Tier 1 Standard outputs: Released when they are in your approved output manifest, consistent with consent, with small counts suppressed.
- Cohort counts by criteria
- GWAS and QTL summary statistics
- Effect sizes and survival results
- Correlation matrices and signatures
- Figures and aggregate reports
- Tier 2 Reviewed outputs: Released after a disclosure review, because they carry more re-identification risk.
- Model evaluation in place
- Feature-level embeddings
- Results from small or rare cohorts
- Tier 3 By exception: Only under the model weights policy, with the data holder’s written approval.
- Trained model weights
- Fine-tuned checkpoints and adapters
- Never Never released: Nothing that could rebuild or stand in for the dataset.
- Row-level or sample-level records
- Synthetic or derivative datasets
Every output comes with its context: the datasheet version, the consent codes it ran under (GA4GH DUO), the pipeline image fingerprint, the output manifest and the audit record. Drop it straight into a methods section.
For AI and ML teams
Building models?
Bring your model to the data. Validate, fine-tune and benchmark it in place, and take out the performance. Whether trained weights can leave depends on the dataset’s consent and its security posture. Every card tells you before you ask.
You own your results.
The data holder keeps the dataset. LooporaData never receives a copy.
Need a count?
Not technical? Ask us for a count.
We return aggregate numbers, and the dataset is accessed in place.