Accelerated variant curation from scientific literature using biomedical text mining.

Area of Science:

• Bioinformatics
• Genomics
• Computational Biology

Background:

• Biological databases rely on biocuration to standardize data, but manual curation is time-consuming.
• Automation in biocuration is increasing, yet significant manual effort remains.
• Accurate extraction of genomic variant locations from literature is crucial for database updates.

Purpose of the Study:

• To develop an automated method for extracting variant genomic locations from C. elegans research papers.
• To improve the efficiency and accuracy of data collection for the WormBase database.
• To supplement manual biocuration efforts by identifying data missed during manual review.

Main Methods:

• A hybrid approach integrating regular expressions, Named Entity Recognition (NER) using BERT (Bidirectional Encoder Representations from Transformers), and bag-of-words models.
• Application of the model to extract gene-mutation matches from 100 C. elegans research papers.
• Evaluation of the model's precision in identifying genomic variant locations.

Main Results:

• The hybrid model achieved a precision of 82.59% for gene-mutation matches.
• The automated method successfully extracted variant genomic locations from the selected papers.
• The model identified some data that was not captured during manual curation, demonstrating its potential to enhance data completeness.

Conclusions:

• The developed hybrid approach offers an efficient and accurate method for extracting variant genomic locations from scientific literature.
• This automation can significantly speed up the biocuration process for biological databases like WormBase.
• The model's ability to recover missed data highlights its value in improving the comprehensiveness of biological databases.