Synthetic cross-sectional demo

This cross-sectional dataset lets reviewers try autoimmune research workflows. It is independent of real study data. Participants, measurements, effects, batches and missingness are independently simulated, not copied, anonymized, or fitted to patient-level study records.

The Synthetic longitudinal demo remains available for longitudinal demonstrations.

Synthetic identifiers

The numeric part of each synthetic patient ID is also its ClarID subject ID. Prefixes are readable labels; they do not restart the numeric counter.

Fixture	Source patient IDs	ClarID subject numbers
Default	SYN001–SYN030, HC031–HC038	1–38
Private synthetic fixture	PRIV10001	10001
Large synthetic fixture	SYN20001–SYN20030, HC20031–HC20038	20001–20038
Autoimmune demo	DEMOAI30001–DEMOAI30090	30001–30090

Sample IDs retain the source patient ID and their visit suffix. The ClarID generator reads the numeric suffix rather than assigning numbers by row order, and rejects duplicate numbers or IDs outside each fixture's reserved range. These conventions apply only to synthetic fixtures; imported studies retain their source identifiers.

Older synthetic cohorts, projects and runs are not migrated after renumbering. Prepare the synthetic query stores again and create new selections. Regenerate subject sidecars with scripts/generate_synthetic_clarids.py and clarid-tools 0.04 after changing a fixture population. For a new large fixture, pass a fresh directory using --large-dataset-root; the generator refuses to overwrite one.

Contents

Component	Synthetic content
Clinical	90 fictional participants: 30 RA, 30 SLE, 30 healthy controls; one baseline sample each
RNA	1,000 genes, 90 samples; positive fractional expected-count-like values
Methylation	128 fictional CpG features, 84 samples; M-values
Immune cells	Six cell measurements, 81 samples
Serology	Six quantitative markers, 87 samples

Dashboard summarises RSEM expected-count matrices without rounding to integers: mean expected counts per sample, mean genes with counts above zero, and the five genes with the largest total expected counts. These are descriptive summaries, not expression filtering or differential-expression results. On-demand summaries are limited to matrices up to 64 MiB and cached in memory until the file changes. Larger matrices or matrices with invalid count values show unavailable summaries, not zero counts. No extra DuckDB import is required.

Assays share participant identifiers. Different participants lack different assays; immune-cell and serology matrices also contain some missing values. Age, sex, batch and fictional recruitment center are provided in assay metadata. Group sizes and technical batches are deliberately balanced for demonstration. No treatment history or repeated visits are simulated.

Try it in the app

Select Synthetic cross-sectional demo in Datasets and apply the workspace. If its query store is not ready, an administrator must prepare it in Data Administration.

1. Inspect Dashboard's clinical, assay and multimodal coverage views.
2. Ask Query to list patients with Systemic Lupus Erythematosus. Save that cohort; save a healthy-control cohort too.
3. In Analysis, select those A/B cohorts and the synthetic RNA assay. Run limma-voom, not DESeq2: this assay is labelled rsem_expected_counts for compatibility with the expected-count workflow.
4. Inspect the methylation panel using PCA or a cohort comparison. Compare immune-cell and quantitative-serology measurements between the same groups.
5. Check sample counts: assay coverage means some analyses include fewer than 30 participants per group.

Selected interferon-labelled genes have a planted SLE-group increase and selected inflammation-labelled genes have a planted RA-group increase. Some methylation features and quantitative markers have artificial group effects. A shared participant component creates limited cross-modal correlation; RNA and methylation also have batch effects. simulation_truth.json records feature-level planted effects. These are engineering checks, not biological findings or a reproduction of published study results. No exact significance threshold is guaranteed.

The RNA values were not produced by RSEM. Fictional CpG identifiers cannot be used for genomic annotation. The small gene panel is not a transcriptome-wide pathway-enrichment benchmark. Cell fractions are separate measurements, not a complete partition of blood cells. Marker names do not make their simulated distributions clinically realistic.

Installation and regeneration

The source checkout and Docker image include datasets/synthetic-cross-sectional-demo/. A fresh runtime lists it automatically; existing registries are not silently changed. An administrator can register its directory in Data Administration and prepare its query store.

To generate an independent copy in a new directory:

python3 scripts/generate_synthetic_autoimmune_demo.py --output /tmp/autoimmune-demo

The generator uses a fixed seed and the vendored public HGNC identifier list. It does not read study data, runtime configuration, or network resources. It refuses to overwrite an existing output directory. Share a clean installation and this synthetic dataset, not a copy of an existing research runtime or its backups.