CASBERT

CASBERT converts composite semantic annotations of biosimulation model entities (e.g., reactions, variables, components) into BERT-derived embeddings to enable text-based retrieval of RDF/ontology-annotated models and improve their findability and reuse.


Key Features:

  • Semantic Annotation Utilization: Processes composite semantic annotations expressed with ontologies and encoded in RDF, covering entities such as reactions, variables, and components and addressing annotations that are typically queried via SPARQL.
  • Text-Based Information Retrieval: Supports retrieval using natural-language queries mapped to embedding space as an alternative to ontology- or SPARQL-based queries.
  • Integration of BERT: Adapts Bidirectional Encoder Representations from Transformers (BERT) to convert composite annotations into dense entity embeddings.
  • Embedding Storage and Indexing: Stores and indexes entity embeddings in a list structure that facilitates addition, modification, and insertion operations for retrieval.
  • Query Embedding and Similarity Ranking: Transforms queries into embeddings and ranks entities by similarity between query embeddings and stored entity embeddings.
  • Scalability and Efficiency: Employs an embedding list structure intended to support efficient operations suitable for large-scale repositories.

Scientific Applications:

  • Model discovery: Locate biosimulation models and annotated model entities relevant to specific biological or biochemical concepts.
  • Model composition and reuse: Identify semantically matching components, variables, or reactions to support composition and reuse of model parts.
  • Model reproduction and curation: Assist reproduction and curation workflows by retrieving annotated entities needed for replication and annotation quality control.

Methodology:

Composite annotations are converted into entity embeddings using BERT, stored in an indexed list, and queries are transformed into embeddings and matched against stored embeddings by similarity; validation used datasets from the Physiome Model Repository and a static export of the BioModels database with performance assessed by Mean Average Precision (MAP) and Mean Reciprocal Rank (MRR) against bag-of-words baselines.

Topics

Details

License:
GPL-3.0
Cost:
Free of charge
Tool Type:
library
Operating Systems:
Mac, Linux, Windows
Programming Languages:
Python
Added:
3/30/2023
Last Updated:
11/24/2024

Operations

Publications

Munarko Y, Rampadarath A, Nickerson DP. CASBERT: BERT-based retrieval for compositely annotated biosimulation model entities. Frontiers in Bioinformatics. 2023;3. doi:10.3389/fbinf.2023.1107467. PMID:36865672. PMCID:PMC9971925.