LSD600: the first corpus of biomedical abstracts annotated with lifestyle–disease relations

In this study, we introduce LSD600, the first corpus specifically focused on lifestyle factor-disease relations. LSD600 consists of 600 abstracts annotated with LSF-disease relations, encompassing 1897 relations covering eight different relation types.

The annotated LSF entities in these relations are pre-annotate using a comprehensive LSF classification (Nourani et al., 2024) and cover  a wide spectrum of lifestyle factors belonging to nine categories.

We have used LSD600 to train a transformer-based model on the multi-label LSF-disease RE task.

Included Files:

1. LSD600 corpus in BRAT format
2. LSD600 corpus annotation guidelines
3. LSD600 Metadata 
4. BRAT config files: Configuration files for loading and visualizing the LSD600 corpus in the BRAT annotation tool.
5. Trained best model for LSF-Disease relation extraction: 
   
   
   

LSD600 Metadata: 

         1. Columns in the Consolidated Relations Dataset :

1. • Disease_Entity: The disease-related term annotated in the text.
   • LSF_Entity: The lifestyle factor (LSF) entity annotated in the text
   • Relationship_Type: The type of relationship between the Disease and LSF entities.
   • Publication_ID: The PubMed ID 
   • Data_Set: (Train/Dev/Test)
   • Disease_BRAT_ID: The BRAT tool’s unique identifier for the Disease entity in the dataset.
   • LSF_BRAT_ID: The BRAT tool’s unique identifier for the LSF entity in the dataset.
   • Disease_Entity_Span: The character span (start and end positions) of the Disease entity in the text.
   • LSF_Entity_Span: The character span (start and end positions) of the LSF entity in the text.
   • Abstract_Text: The full abstract text from which the relationship was extracted.

2. Columns in the LSD600 metadata:

   • Publication_ID: The PubMed ID.
   • Article_Title
   • Journal_Name
   • Publication_Year
   • Data_Set: (Train/Dev/Test)
   • Lifestyle_Factor_Count: The number of lifestyle factors (LSFs) identified in the abstract.
   • Disease_Count: The number of disease entities identified in the abstract.
   • Relations_Count: The total number of relationships between lifestyle factors and diseases annotated in the abstract.
   • Abstract_Text
   • Unique_Lifestyle_Factors: The list of unique lifestyle factors annotated in the abstract.
   • Unique_Diseases: The list of unique diseases annotated in the abstract.