A machine learning-enabled open biodata resource inventory from the scientific literature.

Area of Science:

• Bioinformatics
• Data Science
• Life Sciences

Background:

• Modern biological research relies heavily on biodata resources for data archiving, curation, and analysis.
• Sustained funding for this global infrastructure of biodata resources presents a significant challenge.
• The Global Biodata Coalition (GBC) aims to develop sustainable funding strategies for these essential resources.

Purpose of the Study:

• To develop an automated method for creating a comprehensive global inventory of biodata resources.
• To overcome limitations of existing registries that require manual curation and self-registration.
• To enable periodic updates to the inventory with minimal human intervention.

Main Methods:

• Utilized machine learning-enabled natural language processing (NLP) on Europe PMC publication data.
• Fine-tuned Bidirectional Encoder Representations from Transformers (BERT) models for resource identification and name prediction.
• Incorporated manual review for low-confidence predictions and duplicate resolution, supplemented by article metadata for funder and geolocation data.

Main Results:

• Generated an inventory of 3112 unique biodata resources from articles published between 2011 and 2021.
• The NLP approach successfully identified biodata resources from publication titles and abstracts.
• Developed automated pipelines for reproducible inventory generation, with all code and data released under permissive licenses.

Conclusions:

• The automated NLP-driven approach provides an efficient and scalable method for cataloging biodata resources.
• This inventory aids the Global Biodata Coalition in assessing the scope and support of the global biodata infrastructure.
• The open release of code and data promotes reuse and supports the sustainability of vital biodata resources for biological research.