CASBERT: BERT-based retrieval for compositely annotated biosimulation model entities.

Area of Science:

• Computational Biology
• Bioinformatics
• Systems Biology

Background:

• FAIR data principles are crucial for biosimulation models, requiring comprehensive entity descriptions.
• Current semantic annotation methods using RDF and SPARQL are complex for many users.
• Accessible model entity retrieval is needed to enhance model reuse and reproducibility.

Purpose of the Study:

• To propose CASBERT, a user-friendly, text-based information retrieval system for biosimulation model entities.
• To enable efficient searching and discovery of model components without requiring deep ontology or SPARQL knowledge.
• To improve the FAIRness and accessibility of biosimulation models in repositories.

Main Methods:

• CASBERT adapts Bidirectional Encoder Representations from Transformers (BERT) to create entity embeddings from composite annotations.
• Entity embeddings are stored in a list for efficient comparison with query embeddings.
• A text-based query is converted into an embedding and compared against stored entity embeddings to rank relevant entities.

Main Results:

• CASBERT demonstrated superior performance compared to traditional bag-of-words methods in retrieving relevant entities.
• The system was tested on datasets from the Physiome Model Repository and BioModels database.
• The list-based structure allows for efficient addition, modification, and insertion of entity embeddings, supporting search engine implementation.

Conclusions:

• CASBERT offers an accessible and effective solution for searching biosimulation model entities, overcoming the limitations of SPARQL.
• The approach enhances model discoverability and facilitates reuse, curation, and composition.
• This method significantly improves the FAIRness of biosimulation models for a broader scientific audience.