Evaluate what each omic layer adds
omicau aligns molecular assays and outcomes, compares single-layer and fusion predictors, and writes an auditable report. Use the command line for scripted analyses or the local browser interface for guided setup.
Workflow
A run starts with modality matrices, an endpoint, and an independent grouping unit. It records alignment and data-quality checks, fits reference and fusion models under shared partitions, and reports prediction, calibration, control, and layer-level results in one local output directory.
| Stage | Output for the analyst |
|---|---|
| Prepare | Aligned samples, modality availability, feature counts, missingness, and available batch information |
| Evaluate | Single-layer, fusion, and leave-one-layer-out model metrics under shared group-aware splits |
| Interpret | Modality ledger, paired gain intervals, calibration, controls, and model card |
| Reproduce | Resolved configuration, input provenance, machine-readable audit, and runtime record |
Analysis checks
Omicau keeps independent groups together across assessment splits and fits learned preprocessing within training data. It records diagnostic and control results beside predictive estimates so that users can inspect the conditions of an individual run.
- Input review: sample alignment, missingness, group structure, and available batch labels.
- Model review: shared partitions, single-layer baselines, fusion comparisons, and calibration.
- Control review: eligible target and feature controls with their exact scope recorded in the audit.
Quick start
Download the v0.5.2 wheel from the GitHub release, create the bundled synthetic example, run it, verify its provenance, and inspect the resulting report. Python 3.10 or later is required.
python -m pip install ./omicau-0.5.2-py3-none-any.whl
omicau bootstrap --dataset mock --out-dir demo
omicau run --config demo/config.json --cores 8
omicau verify --config demo/config.json --audit demo/run/audit.jsonOpen demo/run/report.html after the run. Choose a core count appropriate to the available memory and workload.
The release also provides a portable Windows x86-64 ZIP, Linux x86-64 AppImage, and macOS Apple Silicon disk image. As checked on 23 September 2026, PyPI lists v0.4.0; an unpinned pip install omicau does not install v0.5.2.
Inputs and outputs
Provide one numeric matrix for each omics layer and a clinical table containing the endpoint. Include a group identifier when related samples must remain in the same evaluation split, and a batch variable when it is available.
| Input | Purpose |
|---|---|
| Omics matrices | Numeric features for each assayed layer |
| Clinical table | Sample identifier, endpoint, and optional group and batch columns |
| Configuration file | Paths, endpoint definition, evaluation settings, and output location |
Each run writes an interactive report.html, machine-readable audit.json, and tabular exports including model metrics and modality-level results. Use the configuration and run outputs together when reproducing an analysis.
Reproducibility and release materials
Each run records its resolved configuration, input identity, group-aware split, diagnostics, model outputs, and runtime. The published evaluation archive contains post-execution aggregate results and reproduction materials. The earlier frozen suite documents the pre-execution protocol. Neither archive is invoked by ordinary command-line or browser runs, and the evaluation does not establish universal software superiority or clinical utility.
| Resource | Use |
|---|---|
| Omicau v0.5.2 release | Wheel, source archive, Windows ZIP, Linux AppImage, and macOS arm64 disk image |
| GitHub repository | Source and usage documentation |
| Published evaluation archive | Post-execution aggregate results and reproduction materials |
| Historical frozen suite | Pre-execution protocol and source-registry context |
omicau verify --config config.json --audit run/audit.jsonResearch-use limitation
Interpret a predictive result for the cohort, endpoint, grouping, and assays that generated it. Follow-up biological and clinical questions need study designs suited to those purposes.
Links
Article: work in progress.