BioTriplex

This repository contains the code and instructions to replicate the baseline results for the BioTriplex dataset, a biomedical Relation Extraction (RE) and Named Entity Recognition (NER) dataset. The dataset includes expert annotations on PubMedCentral full-text articles, focusing on Gene, Disease and the relations between them.

This repo provides:

• the BioTriplex corpus (/data/BioTriplex Corpus) (full-text gene, disease and relation annotations),
• preprocessed splits for model training (/data/Preprocessed BioTriplex),
• BERT-based baselines, and
• the LLaMA-based experiments, via the llama-rec framework.

Repository Structure

BioTriplex-results/
├── LICENSE                     # Code license (MIT)
├── README.md                   # This file
├── code
│   ├── README.md               # Additional code-level documentation (if provided)
│   ├── bert                    # BERT-based RE baselines on BioTriplex
│   │   ├── __init__.py
│   │   ├── biotriplex_qakshot_dataset.py
│   │   └── finetune_biotriplex_bert.py
│   ├── llama-rec               # LLaMA-based RE experiments (llama-rec framework)
│   │   ├── CODE_OF_CONDUCT.md
│   │   ├── CONTRIBUTING.md
│   │   ├── README.md
│   │   ├── UPDATES.md
│   │   ├── dev_requirements.txt
│   │   ├── docs
│   │   ├── pyproject.toml
│   │   ├── recipes
│   │   ├── requirements.txt
│   │   ├── src
│   │   └── tools
│   ├── dygixepp  # Have not been uploaded yet
│   └── API-models # LLMs and PubTator API-based RE and NER experiments
└── data
    ├── BioTriplex Corpus   # The annotated BioTriplex full-text articles
    │   ├── 2275485
    │   ├── 2807459
    │   ├── ...
    │   └── 9992796
    ├── README.md              # Dataset description / format details
    └── Preprocessed BioTriplex   # Train/val/test splits for modelling
        ├── splits.json
        ├── test_para.txt
        ├── test_shorter.txt
        ├── train_para.txt
        ├── train_shorter.txt
        ├── val_para.txt
        └── val_shorter.txt