Web Interface

Web Interface

RareSim includes a browser-based interface for running patient diagnosis interactively. It consists of a Vue 3 frontend and a FastAPI backend.

Running the Interface

Terminal 1 — backend:

Create a .env file at the project root with RARESIM_ROOT=/path/to/RareSim (see Installation), then: ```bash uvicorn raresim_api.main:app --reload --port 8000 ```

Terminal 2 — frontend: ```bash cd packages/raresim-frontend npm install npm run dev ``` Then open http://localhost:3000.

Input Panel

The left panel handles patient input and method selection.

Phenotype Search

A searchbar at the top lets you find HPO terms by name (e.g. typing "ataxia" returns all matching phenotypes). Each result has two buttons:

• + Include — adds the term to the patient's HPO term list
• − Exclude — marks the term as excluded; excluded terms are filtered out before diagnosis runs

Excluded terms are sent to the backend as excluded_hpo_terms and removed from the patient's HPO set before any similarity method runs.

Input Modes

• HPO Terms — paste HPO IDs directly (e.g. HP:0001251, HP:0000545). Terms are parsed and displayed as tags.
• Raw Text — paste clinical notes. Use the Extract bar to map text to HPO terms via dictionary lookup, FastHPOCR, GPT-4o-mini, PhenoBrain, or BioNER.

Similarity Methods

Select one or more methods to run. Available methods:

Method	Badge	Notes
Resnik BMA	IC	Semantic similarity using information content
Lin BMA	IC	Lin's normalized semantic similarity
JC BMA	IC	Jiang-Conrath semantic similarity
Jaccard	set	Set overlap: intersection / union
Dice	set	Set overlap: 2 × intersection / (A + B)
TF-IDF (HPO)	txt	TF-IDF over HPO term presence — HPO-terms mode only
TF-IDF (HPO Labels)	txt	TF-IDF over HPO label text — HPO-terms mode only
TF-IDF (Text)	txt	TF-IDF over raw clinical text — raw-text mode only
TF-IDF (Hybrid)	txt	Patient HPO labels vs. disease description — both modes
Transformer	emb	Sentence transformer embeddings
LLM	llm	GPT-based ranking
HPO2Vec+	emb	Node2Vec embeddings on enriched HPO graph
Autoencoder	nn	Denoising autoencoder latent space similarity.

Top-K

Slider to control how many results are returned per method (5, 10, 15, or 20).

Results Panel

The right panel shows diagnosis results after running.

Method Filter

When multiple methods are selected, filter buttons appear above the results list —> one per method plus an "All" option. Clicking a method shows only that method's top-K results

Result Cards

Each card shows:

• Rank, disease label, disease ID
• Method used (shown as a badge)
• Score with a visual bar
• Expandable detail section with shared phenotypes and top term matches

Save Patient

After running a diagnosis, click Save Patient to save the session to disk. Choose the format before saving:

• JSON — saves HPO terms, raw text, methods used, and full results
• Phenopacket — saves HPO terms in GA4GH phenopacket format with results in metadata

Files are saved to outputs/webapp/patient_profiles/ and can be retrieved via GET /api/patients.

API Endpoints

Method	Path	Description
POST	/api/extract	Extract HPO terms from clinical text
POST	/api/diagnose	Run similarity diagnosis
GET	/api/hpo/search?q=	Search HPO terms by label
POST	/api/patients/save	Save patient session to disk
GET	/api/health	Health check