Using Machine Learning to Collect and Facilitate Remote Access to Biomedical Databases: Development of the Biomedical...

Area of Science:

• Bioinformatics
• Computational Biology
• Data Science

Background:

• Existing biomedical literature repositories lack mechanisms for locating and accessing specific biomedical databases.
• This limitation hinders efficient data retrieval and utilization in scientific research.

Purpose of the Study:

• To develop the Biomedical Database Inventory (BiDI), a novel repository for discovering and accessing biomedical databases.
• BiDI aims to automatically extract database information from scientific literature, providing a centralized index and seamless access.

Main Methods:

• An ensemble of deep learning and natural language processing models was employed for database mention extraction.
• A dataset of 1242 annotated articles was used for training, incorporating transfer learning techniques.
• The models achieved high performance in detecting database publications and extracting associated weblinks.

Main Results:

• The system achieved an F1 score of 0.929 for database detection and 0.908 for weblink extraction.
• Over 10,000 unique biomedical databases were identified by applying the model to PubMed and PubMed Central.
• Use cases demonstrated BiDI's utility for 'omics' and COVID-19 research, highlighting its ability to filter irrelevant links.

Conclusions:

• BiDI significantly improves access to biomedical resources, supporting data-driven research and scientific endeavors.
• The repository is openly accessible online and will be continuously updated via an automated text processing pipeline.
• The methodology is adaptable for creating similar repositories across various scientific domains.