RelCurator: a text mining-based curation system for extracting gene-phenotype relationships specific to...

Area of Science:

• Computational Biology
• Bioinformatics
• Genetics

Background:

• Identifying gene-phenotype relationships is crucial for precision medicine.
• Most gene-phenotype data is unstructured and found within biomedical literature.
• Automated extraction of this data is needed to advance medical genetics.

Purpose of the Study:

• To develop RelCurator, a system for extracting gene-phenotype relationships from PubMed articles.
• To predict gene-phenotype relationships for neurodegenerative disorders using deep learning.
• To provide a user-friendly interface for curators to access and utilize this information.

Main Methods:

• Developed a deep learning model using Bidirectional Gated Recurrent Unit (BiGRU) networks and BioWordVec embeddings.
• Trained the model on over 130,000 labeled PubMed sentences related to neurodegenerative disorders.
• Compared the BiGRU model's performance against BERT, SVM, and simple RNN models.

Main Results:

• The RelCurator system achieved a high F1-score of 0.96 in predicting gene-phenotype relationships.
• Demonstrated effectiveness in real-world curation scenarios, identifying potential new causative and associated genes.
• Outperformed other models including BERT, SVM, and simple RNN.

Conclusions:

• RelCurator effectively identifies gene-phenotype relationships, including novel associations for neurodegenerative disorders.
• The system offers a user-friendly interface and deep learning-based insights for curators.
• Represents a significant advancement in the automated curation of gene-phenotype relationships.