README.md

---

dataset_info:
  features:
  - name: NofPmids
    dtype: float64
  - name: NofSnps
    dtype: float64
  - name: associationType
    dtype: string
  - name: diseaseId
    dtype: string
  - name: diseaseName
    dtype: string
  - name: diseaseType
    dtype: string
  - name: disease_mention
    dtype: string
  - name: geneId
    dtype: string
  - name: geneSymbol
    dtype: string
  - name: gene_mention
    dtype: string
  - name: originalSource
    dtype: string
  - name: pmid
    dtype: int64
  - name: raw_sentence
    dtype: string
  - name: score
    dtype: float64
  - name: sentence
    dtype: string
  - name: source
    dtype: string
  splits:
  - name: train
    num_bytes: 1907978
    num_examples: 4000
  - name: validation
    num_bytes: 1134075
    num_examples: 2400
  - name: test
    num_bytes: 756401
    num_examples: 1600
  download_size: 2013053
  dataset_size: 3798454
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
  - split: validation
    path: data/validation-*
  - split: test
    path: data/test-*
task_categories:
- text-classification
language:
- en
tags:
- medical
- biology
- relation-classification
- relation-extraction
- gene
- disease
pretty_name: GDA
size_categories:
- 1K<n<10K
---

# Dataset Card for GDA

## Dataset Description

- **Repository:** https://github.com/EsmaeilNourani/Deep-GDAE
- **Paper:** [Association extraction from biomedical literature based on representation and transfer learning](https://www.sciencedirect.com/science/article/abs/pii/S0022519319304813)

#### Dataset Summary

<!-- Provide a quick summary of the dataset. -->
**GDA Dataset Summary:**

Nourani and Reshadata (2020) developed a dataset called GDA corpus as a sentence-level evaluation dataset for extracting the association between genes and diseases based on some efficient databases.
They used DisGeNET, a database of Gene-Disease-Associations (GDAs), and PubTator to retrieve biomedical texts (PubMed abstracts).
Using PubTator, they found all the PMIDs containing at least one gene and disease name. 
Samples of the true class were extracted from DisGeNET, considering only curated associations.
For the creation of non-associated or false samples, a systematic three-step filtering process was used to ensure high-quality annotations. 
This included the exclusion of known associations from DisGeNET and CTD, as well as a linguistic filtering step to remove sentences that linguistically imply an association.
GDA was constructed automatically and contains 8000 sentences with 1904 and 3635 unique diseases and genes respectively.


### Languages
The language in the dataset is English.

## Dataset Structure
<!-- This section provides a description of the dataset fields, and additional information about the dataset structure such as criteria used to create the splits, relationships between data points, etc. -->
### Dataset Instances
An example of 'train' looks as follows:
```json
{
  "NofPmids": 2.0,
  "NofSnps": 0.0,
  "associationType": "Biomarker",
  "diseaseId": "C0043459",
  "diseaseName": "Zellweger Syndrome",
  "diseaseType": "disease",
  "disease_mention": "H-02",
  "geneId": "5194",
  "geneSymbol": "PEX13",
  "gene_mention": "PEX13",
  "originalSource": "CTD_human",
  "pmid": 10332040,
  "raw_sentence": "We now have evidence that the complete human cDNA encoding Pex13p, an SH3 protein of a docking factor for the peroxisome targeting signal 1 receptor (Pex5p), rescues peroxisomal matrix protein import and its assembly in fibroblasts from PBD patients of complementation group H. In addition, we detected mutations on the human PEX13 cDNA in two patients of group H. A severe phenotype of a ZS patient (H-02) was homozygous for a nonsense mutation, W234ter, which results in the loss of not only the SH3 domain but also the putative transmembrane domain of Pex13p.",
  "score": 0.400549453568426,
  "sentence": "We now have evidence that the complete human cDNA encoding Pex13p, an SH3 protein of a docking factor for the peroxisome targeting signal 1 receptor (Pex5p), rescues peroxisomal matrix protein import and its assembly in fibroblasts from PBD patients of complementation group H. In addition, we detected mutations on the human <span class=\"gene\" id=\"10332040-3-326-331\">PEX13</span> cDNA in two patients of group H. A severe phenotype of a <span class=\"disease\" id=\"10332040-3-389-391\">ZS</span> patient (<span class=\"disease\" id=\"10332040-3-401-405\">H-02</span>) was homozygous for a nonsense mutation, W234ter, which results in the loss of not only the SH3 domain but also the putative transmembrane domain of Pex13p.",
  "source": "CTD_human;ORPHANET"
}
```

### Data Fields

Here's the Data Fields section for the GDA corpus based on the dataset features provided:

- `NofPmids`: the number of PubMed IDs related to the gene-disease association, stored as a `float64` feature.
- `NofSnps`: the number of single nucleotide polymorphisms (SNPs) related to the gene-disease association, stored as a `float64` feature.
- `associationType`: the type of association (e.g., Negative, Biomarker, Therapeutic) between the gene and the disease, a `string` feature.
- `diseaseId`: the unique identifier for the disease discussed, a `string` feature.
- `diseaseName`: the name of the disease, a `string` feature.
- `diseaseType`: the type of the disease (e.g., disease, group, phenotype), a `string` feature.
- `disease_mention`: the specific mention of the disease within the source text, a `string` feature.
- `geneId`: the unique identifier for the gene discussed, a `string` feature.
- `geneSymbol`: the symbol representing the gene, a `string` feature.
- `gene_mention`: the specific mention of the gene within the source text, a `string` feature.
- `originalSource`: the original source, a `string` feature.
- `pmid`: the PubMed ID associated with the citation from which the sentence is derived, an `int64` feature.
- `raw_sentence`: the original sentence from the source document, a `string` feature.
- `score`: a score reflecting the confidence or relevance of the association between the gene and the disease, stored as a `float64` feature.
- `sentence`: the sentence with span annotation, a `string` feature.
- `source`: the database or repository from which the association data was taken, a `string` feature.

## Citation
<!-- If there is a paper or blog post introducing the dataset, the APA and Bibtex information for that should go in this section. -->

**BibTeX:**
```
@article{NOURANI2020110112,
title = {Association extraction from biomedical literature based on representation and transfer learning},
journal = {Journal of Theoretical Biology},
volume = {488},
pages = {110112},
year = {2020},
issn = {0022-5193},
doi = {https://doi.org/10.1016/j.jtbi.2019.110112},
url = {https://www.sciencedirect.com/science/article/pii/S0022519319304813},
author = {Esmaeil Nourani and Vahideh Reshadat},
keywords = {Gene-Disease Association Extraction, Attention Mechanism, BioBERT}
}
```

**APA:**
- Nourani, E., & Reshadat, V. (2020). Association extraction from biomedical literature based on representation and transfer learning. Journal of Theoretical Biology, 488, 110112. https://doi.org/10.1016/j.jtbi.2019.110112

## Dataset Card Authors
[@phucdev](https://github.com/phucdev)