---
title: BioBERT outputs & interpretation
description: 'Entity spans, relation triples with evidence sentences, and QA answers with confidence scores. BioBERT is deprecated and chat cannot call it.'
---

BioBERT returns extracted entities, relationship triples, or answers — each with a type and a confidence score, exportable as annotated text, CSV, or JSON. Use these to feed knowledge graphs, prioritize targets, or support literature-review workflows.

> **Not available in chat**

> BioBERT is deprecated on the platform and not registered with the orchestrator, so chat cannot call it. A current LLM performs BioBERT's entity extraction and relevance ranking natively over text retrieved with [PubMed](/docs/pubmed); for curated identifiers an LLM must not invent (NCBI Gene IDs, MeSH terms), use PubTator3. Existing BioBERT job results remain viewable.

## Output formats

## Named Entity Recognition (NER)

**Annotated text:**
- Entity spans with start/end positions
- Entity type classifications
- Confidence scores per entity

**CSV export:**
```csv
text,entity,type,start,end,confidence
"BRCA1 mutation...",BRCA1,GENE,0,5,0.98
```

## Relation Extraction (RE)

**Relationship triples:**
- Subject entity (with type)
- Relationship type
- Object entity (with type)
- Evidence sentence
- Confidence score

**CSV export:**
```csv
subject,subject_type,relation,object,object_type,evidence,confidence
Aspirin,DRUG,inhibits,COX-2,PROTEIN,"Aspirin inhibits...",0.94
```

## Question Answering (QA)

**Answer extraction:**
- Extracted answer text
- Answer span in source
- Supporting context
- Confidence score

## Entity types

| Type | Description | Examples |
|------|-------------|----------|
| GENE | Genes and proteins | BRCA1, p53, insulin |
| DISEASE | Diseases and conditions | cancer, diabetes |
| CHEMICAL | Chemicals and drugs | aspirin, glucose |
| SPECIES | Organisms | human, mouse, E. coli |
| MUTATION | Genetic variants | V600E, c.248T>C |
| CELL | Cell types | T cells, neurons |

## Confidence interpretation

- **High (>0.9)**: Very confident prediction
- **Medium (0.7-0.9)**: Confident, suitable for most uses
- **Low (0.5-0.7)**: Less certain, review recommended
- **Very low (<0.5)**: Not output by default

## Canvas Visualization

When a BioBERT job completes, a result card appears on the canvas.

- **Card size**: 400x300px
- **Border color**: Brown (#795548)
- **Visualization**: Entity network showing extracted entities and relationships

### Tier 2 Metrics

- Entity count by type (genes, proteins, diseases, compounds)
- Relation count
- Entity types extracted
- Confidence scores per entity

## Read next