Dataset of miRNA-disease relations extracted from textual data using transformer-based neural networks

Area of Science:

• Biomedical informatics
• Computational biology
• Genomics

Background:

• MicroRNAs (miRNAs) are crucial regulators of cellular functions, and their dysregulation is implicated in diseases like cancer and neurodegenerative disorders.
• Identifying miRNA-disease associations from vast biomedical literature is challenging due to manual extraction's time and labor intensity.

Purpose of the Study:

• To develop and evaluate a deep learning-based text mining approach for extracting normalized miRNA-disease associations.
• To create an extended training corpus using distant supervision from multiple databases.

Main Methods:

• A deep learning model was trained on a novel corpus augmented with distant supervision.
• The approach was quantitatively evaluated for its ability to detect miRNA-disease associations.
• The model's applicability was demonstrated by extracting associations from PubMed and PubMed Central.

Main Results:

• The developed workflow achieved a 98% area under the receiver operator characteristic curve on a test set.
• The approach successfully extracted novel miRNA-disease associations, particularly for neurodegenerative diseases.
• These extracted associations were not previously present in public databases.

Conclusions:

• The deep learning text mining approach effectively and accurately extracts miRNA-disease associations from biomedical literature.
• This method offers a valuable tool for researchers to discover new disease-related miRNAs, accelerating biomedical research.
• The findings highlight the potential of AI in uncovering complex biological relationships from unstructured data.