COMMODAR

COMMODAR extracts context-specific molecular relations from unstructured biomedical literature to produce structured representations of molecular interactions for computational biology and bioinformatics analyses.

Key Features:

• Multimodal Representations: Combines biomedical domain knowledge with canonical linguistic information to augment features for relation extraction.
• Machine Learning Framework: Employs a machine learning–based framework, including neural networks, to perform relation extraction from text.
• Context-Specific Extraction: Focuses on extracting molecular relations that are specific to biological contexts rather than generic associations.
• Large-Scale Literature Mining: Applied to 14 million PubMed abstracts and produced 9,214 context-specific molecular relations.

Scientific Applications:

• Biological Network Construction: Provides context-specific molecular relations for building and refining biological interaction networks.
• Literature-Scale Knowledge Extraction: Enables extraction of molecular relations from large corpora of biomedical texts for downstream computational analyses.

Methodology:

Multimodal representations combining biomedical domain knowledge and canonical linguistic information; machine learning–based relation extraction using neural networks; large-scale application to PubMed abstracts (14 million) yielding 9,214 extracted context-specific molecular relations.