GDA Dataset Overview

Dataset Description

Dataset Summary

The GDA dataset was developed by Nourani and Reshadata (2020) as a sentence‑level evaluation resource for extracting gene‑disease associations from biomedical literature. It is built on the DisGeNET and PubTator databases and comprises 8,000 sentences covering 1,904 unique diseases and 3,635 unique genes.

Language

The language of the dataset is English.

Dataset Structure

Data Fields

• NofPmids: Number of PubMed IDs associated with the gene‑disease pair, type float64.
• NofSnps: Number of single‑nucleotide polymorphisms (SNPs) linked to the gene‑disease pair, type float64.
• associationType: Type of association between gene and disease (e.g., Negative, Biomarker, Therapeutic), type string.
• diseaseId: Unique identifier for the disease, type string.
• diseaseName: Disease name, type string.
• diseaseType: Category of disease (e.g., disease, group, phenotype), type string.
• disease_mention: Specific mention of the disease in the source text, type string.
• geneId: Unique identifier for the gene, type string.
• geneSymbol: Symbolic representation of the gene, type string.
• gene_mention: Specific mention of the gene in the source text, type string.
• originalSource: Original source, type string.
• pmid: PubMed ID linked to the sentence, type int64.
• raw_sentence: Original sentence from the source document, type string.
• score: Confidence or relevance score for the gene‑disease association, type float64.
• sentence: Sentence with span annotations, type string.
• source: Database or repository providing the association data, type string.

Data Splits

• train: Training set with 4,000 samples, size 1,907,978 bytes.
• validation: Validation set with 2,400 samples, size 1,134,075 bytes.
• test: Test set with 1,600 samples, size 756,401 bytes.

Citation

• 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