DeepKG

DeepKG extracts biomedical knowledge from literature and represents it as knowledge graphs using deep learning and AutoML to support disease mechanism elucidation, drug repurposing, and clinical research.

Key Features:

• Automated Knowledge Graph Construction: Constructs domain-specific knowledge graphs by extracting entities and relations from biomedical texts.
• Cascaded Hybrid Information Extraction Framework: Employs a cascaded hybrid framework to train models that extract 3-tuples (subject-predicate-object) from text data.
• AutoML-based Knowledge Representation (AutoTransX): Integrates AutoTransX, an AutoML algorithm, to optimize knowledge representation and inference from extracted triples.
• Large-scale COVID-19 Knowledge Extraction: Applied to 144,900 full-text COVID-19 articles to generate a knowledge graph comprising 7,980 entities and 43,760 3-tuples, including candidate drug lists and links to animal experimental studies.

Scientific Applications:

• Disease Mechanism Analysis: Supports elucidation of biological mechanisms by integrating entity and relation data into structured knowledge graphs.
• Drug Repurposing: Produces candidate drug lists and relational evidence to inform drug repurposing hypotheses.
• Clinical and Translational Research: Enables literature-scale evidence aggregation to inform clinical studies and experimental validation.

Methodology:

Uses a deep learning-based workflow with a cascaded hybrid information extraction framework for 3-tuple extraction and AutoTransX AutoML for knowledge representation and inference.