Web Application

The Streamlit frontend is a minimal researcher UI over the same core pipeline as the CLI. It computes the FLB factor for one or more pedigrees and produces a downloadable run-provenance manifest for every run — identical to what the CLI emits, because both call PipelineRunner.run_with_manifest through the shared build_config_from_dict helper.

Launch

Install the web extra and start the app:

pip install -e ".[web]"
streamlit run src/heredicalc/apps/web/app.py

The app opens in your browser (default http://localhost:8501).

Parameters

The form is pre-filled with the primary Belman reference case (brca1_belman_latvia_ci5ix, victor):

Widget	Default	Notes
Genetic entity	BRCA1	Free text
Allele frequency	0.0001	Hardy-Weinberg prior
Age bands	30,40,50,60,65,70,80	Comma-separated integers
Incidence source	ci5_ix	Dropdown of registered incidence sources
Phenotype model	hbopc	Dropdown of registered phenotype models
Population	Latvia	Dropdown from the source's list_sources(); value is the source_id
Trait mapper	ci5_ix_hbopc	Dropdown of mappers compatible with the source + phenotype model
RR model	tabular	Dropdown of registered RR models compatible with the selection
CRHF model	lookup	Dropdown of registered CRHF models compatible with the selection

Every dropdown is populated from the registry discovered by build_registry() / discover_all() — never from a hardcoded list — so externally installed plugins appear automatically (see External plugins in the dropdowns).

The model backbone (hazard_model, penetrance_model=victor, liability_assigner, flb_calculator=segregatr, pedigree_format=cool3_tsv) is fixed to the reference defaults and shown in the Fixed model & plugins panel.

Population dropdown

The population is a dropdown filled from the selected incidence source's list_sources(). Each entry shows the full registry name plus study period — so registries that share a prefix stay distinct (e.g. USA, SEER (9 Registries), USA, SEER (9 Registries): Black, and the various Korea, … registries are all separately selectable).

The selection value is the registry's unambiguous source_id, and that source_id is passed as params["population"]. The runner resolves it through find_source_id's exact-ID branch, so there is no substring ambiguity (a name like Korea no longer matches many registries). Changing the incidence source refreshes the list; a source that exposes no populations shows a clear error.

Dependent plugin coupling

The trait-mapper, RR-model, and CRHF-model dropdowns are derived from the current selections using the plugins' declared compatible_with/requires metadata — the same mechanism the registry uses for compatibility validation, so no plugin names are hardcoded:

• Trait mapper declares compatible_with = {incidence_source, phenotype_model}, so its list is restricted to mappers compatible with both the chosen incidence source and phenotype model.
• RR model / CRHF model are present because the fixed penetrance_model (victor) declares requires = {rr_model, crhf_model}; the app derives which sub-plugin kinds to offer from that requires. The built-in tabular/lookup declare no compatible_with, so they are universally compatible; a study-specific RR/CRHF plugin that declares compatible_with = {phenotype_model: [...]} is filtered to the matching phenotype models by the same rule.

When you change the incidence source or phenotype model, the dependent dropdowns update automatically. If no compatible plugin exists for a kind, the app shows a clear error instead of running.

Providing pedigrees

Three ways, in precedence order:

1. Upload one or more COOL3 TSV .ped files (Upload COOL3 TSV pedigrees).
2. Paste a single pedigree into the text area.
3. Load demo — loads the bundled Belman pedigree and resets the parameters to the reference values. Clicking Run FLB then yields FLB ≈ 25.65.

Uploaded files are written to temporary files (the runner takes a path) and cleaned up after the run.

Results and manifests

• A single pedigree shows its FLB prominently; multiple pedigrees produce a results table (pedigree, FLB, status).
• Each successful run offers a <pedigree>.manifest.json download (RunManifest.model_dump_json); with several pedigrees you can also download all manifests as a ZIP.
• An expandable Provenance panel per run shows the manifest's resolved_config and r_session.

External plugins in the dropdowns

Because every dropdown is fed from the discovered registry, study-specific plugins loaded via HEREDICALC_PLUGIN_PATH appear automatically alongside the built-ins — no code change to the app. Point HerediCalc at your plugin directory before launching:

export HEREDICALC_PLUGIN_PATH=~/my_project/plugins
streamlit run src/heredicalc/apps/web/app.py

A new phenotype_model shows up in the phenotype dropdown; a new incidence_source adds its populations; and a trait_mapper/rr_model/crhf_model that declares the appropriate compatible_with is offered automatically whenever its declared selection is active. See External Plugins for the plugin file layout and discovery rules.

Error handling

Parsing and validation are delegated to the pedigree_format plugin. Invalid or empty pedigrees, parse errors, a missing Rscript, or R-side failures are caught and shown as a friendly st.error message — never a stack trace. In a batch, one failing pedigree does not abort the others: its row is marked error with a short reason while the rest complete. A genetic entity with no RR/CRHF data surfaces the same way — a clear error row, not a crash.