BiOnt

BiOnt performs relation extraction in the biomedical domain using deep learning integrated with domain-specific ontologies to identify associations among genes, phenotypes, diseases, and chemical entities.

Key Features:

• Deep Learning Integration: BiOnt uses state-of-the-art deep learning methods for relation extraction.
• Biomedical Ontologies: Integrates four ontology types to represent biomedical entities and guide relation inference.
• Gene Ontology (GO): Represents gene-product functions and annotations used by the model.
• Human Phenotype Ontology (HPO): Encodes human phenotype terms and relations used by the model.
• Human Disease Ontology (DO): Encodes disease concepts and relations used by the model.
• Chemical Entities of Biological Interest (ChEBI): Represents chemical compounds and classifications used by the model.
• Drug-Drug Interaction performance: Achieved a 4.93% improvement in F-score on the DDI corpus.
• Phenotype–Gene relation performance: Achieved a 4.99% increase in F-score on the PGR corpus.
• Chemical–Induced disease performance: Achieved a 2.21% enhancement in F-score on the BC5CDR corpus.

Scientific Applications:

• Drug–Drug interaction extraction: Detects and classifies interactions between drugs from biomedical text.
• Phenotype–Gene relation extraction: Identifies associations between phenotypes and genes from text corpora.
• Chemical–induced disease extraction: Extracts relations linking chemical compounds to induced diseases from literature.
• Ontology-guided relation discovery: Uses ontology knowledge to reveal associations among genes, phenotypes, diseases, and chemicals.

Methodology:

Integrates Gene Ontology (GO), Human Phenotype Ontology (HPO), Human Disease Ontology (DO), and ChEBI into deep learning-based relation extraction models and evaluates performance on the DDI, PGR, and BC5CDR corpora with reported F-score improvements.