Extracting and consolidating a knowledge graph of herb-drug interactions
The kANNa project integrates natural language processing with knowledge modeling, aggregation and reasoning. The overall objective is the development of new methodologies based on Deep Learning in the extraction and representation of relations from text. The scientific objectives of the kANNa project will be achieved by pursuing the following, more specific goals:
Objective 1: Integrate information extraction into the process of monitoring herb-drug interactions from medical literature.
kANNa aims to extract accurate, complete and up to date knowledge about herb-drug interactions from scientific publications. This work will investigate, implement, apply and evaluate a state of the art information extraction pipeline based on Deep Networks and ontologies to detect, classify, integrate and share herb-drug interactions. kANNa will construct a dedicated knowledge base that will allow consumers and medical practitioners to access relevant evidence by reusing and extending existing standards in the field. This collection of relational facts will populate a large knowledge graph where entities are graph nodes and relations are graph edges, a model previously used for extracting adverse drug reactions.
Objective 2: Enhance knowledge acquisition from sparse, incomplete, and unreliable evidence.
In scientific literature, herb-drug interactions are identified through in vitro, in vivo, or clinical studies with varying certainty and often the interaction mechanisms are only hypothesised or unknown. For example, in the case of co-administration of warfarin and St John’s Wort the proposed mechanism is induction of CYP450 isoenzymes by constituents of St. John's Wort, but there is only limited clinical data available. Due to space constraints and their target audience, scientific publications often use implicit information when describing interaction mechanisms, in which case the extracted knowledge base will also be incomplete and must be enhanced through knowledge graph completion. kANNa will investigate methods for enriching the extracted knowledge graph by learning missing relations . This will make it possible to hypothesise interaction mechanisms and analyse herbal drugs with similar interaction patterns.
Objective 3: Provide support for clinical decision making and promote collaboration and reuse over the acquired knowledge base.
kANNa will not only preserve, curate, and enrich the collected herb-drug interaction data but also provide advanced knowledge graph visualisation services . This task will provide the means for an easy comparison between well-known and less-known herb-drug interactions giving broad overviews of the collected information, and guiding researchers to help focus their studies. To increase adoption, kANNa will also facilitate the discussion and collaboration among interested stakeholders including drug companies, regulators, health providers and patients.