Automated Metadata Suggestion During Repository Submission

Area of Science:

• Computational neuroscience
• Bioinformatics
• Data science

Background:

• Knowledge discovery relies on complete data and metadata.
• Manual metadata curation is time-consuming and limited by expertise.
• Diverse research needs create complex metadata schemes, challenging automated annotation.

Purpose of the Study:

• To develop an automated method for improving metadata completeness in informatics resources.
• To reduce the burden on human curators by leveraging existing data.
• To enhance the discoverability of computational models and data.

Main Methods:

• Implemented manually-curated, regular-expression-based rules to parse user-provided text.
• Developed a system to suggest metadata annotations during data entry in the ModelDB resource.
• Analyzed the precision and recall of metadata suggestions from different text sources (abstract, full-text, title).

Main Results:

• Parsing abstracts suggested an average of 6.4 metadata tags per entry with 79% precision.
• Full-text analysis yielded higher recall but lower precision (41%).
• Metadata suggestion effectiveness varied across different annotation categories.

Conclusions:

• Automated metadata suggestion using text parsing can significantly improve informatics resource completeness.
• Prompting data providers with relevant metadata during upload is an effective strategy.
• This approach offers a cost-effective method for other bioinformatics resources to enhance metadata quality.