Towards a foundational representation of potential drug-drug interaction knowledge M Brochhausen, J Schneider, D Malone, PE Empey, WR Hogan, RD Boyce
Jun 15, 2015
Towards a foundational representation of potential drug-drug interaction knowledge
M Brochhausen, J Schneider, D Malone, PE Empey, WR Hogan, RD Boyce
"Addressing gaps in clinically useful evidence on drug-drug interactions"
Goal:• Identify the core components of a new knowledge
representation paradigm that enables better integration of drug-drug interaction evidence – Does the interaction exist?– What potential impact?– How to best manage exposure
Hypothesis:• The new paradigm will result in more complete,
accurate, and current knowledge available for decision support
Important distinctions
• Potential drug-drug interactions (PDDIs) ≠ Drug-drug interactions– A drug-drug interaction is an interaction between
two or more drugs in a actual patient• i.e., some real clinical effect harmful or helpful
– A PDDI involves two or more drugs for which there exists a known interaction potential• i.e., only the potential exists for some real clinical effect• A manageable source of drug-related adverse events
Clinically relevant PDDIs
Van Roon et al. Drug Saf. Int. J. Med. Toxicol. Drug Exp. 28, 1131–1139 (2005).
Three contexts of PDDI knowledge representation
Evidence base: evidence for and against patient risk factors, seriousness of an adverse event incidence
Knowledge base: PDDI assertions pharmacologic/physiological assertions
Reasoning system: ensuring logically consistency of inferred knowledge
with all existing assertions in the knowledge base.
Informational dependencies between Evidence base, Knowledge base and Reasoning system
Motivation
The development of our new semantic resource is motivated by the needs of experts who must search, evaluate, and synthesize PDDI evidence into knowledge claims.
Requirements
Linkage between the semantic model of the evidence base and the knowledge base.
Represent PDDI evidence without inferring the existence of an actual drug-drug interaction. Not all PDDIs are actualized.
Represent drug entities according to the scientific state-of-the-art of pharmacology.
Semantic schema of knowledge base represented in a coding language that allows consistency check.
Potential Drug-Drug Interaction
Definition:
"A potential drug-drug interaction (PDDI) is an information content entity that specifies the possibility of a drug-drug interaction based on either reasonable extrapolation about drug-drug interaction mechanisms or a data item created by clinical studies, clinical observation or physiological experiments."
Related work
Drug Interaction Ontology (DIO) [1] Drug-Drug Interaction Ontology (DINTO) [2]
1 Arikuma et al. BMC Bioinformatics. 9, S11, 2008.2 Herrero-Zazo et al. http://ceur-ws.org/Vol-1114/Session3_Herrero-Zazo.pdf, 2013.
Drug Interaction Ontology (DIO) 1/2
developed with the goal of predicting drug interactions
inspired by both Basic Formal Ontology (BFO) and the NCI Thesaurus (via UMLS)– it is not aligned with either one
No distinction between drug products, their ingredients and the molecules that constitute those ingredients. Each instance of a chemical is a drug, regardless of
dosage or formulation potential to assign incorrect properties
NDF-RT asserts that vancomycin capsules "may treat" bacterial endocarditis and pneumococcal meningitis. However, only intravenous vancomycin can treat those conditions. [1]
Drug Interaction Ontology (DIO) 2/2
1 Hogan WR et al. http://www2.unb.ca/csas/data/ws/icbo2013/papers/research/icbo2013_submission_40.pdf, 2013.
Drug Interaction Ontology (DINTO) 1/2
DINTO is intended "to represent all possible mechanisms that can lead to a drug-drug interaction. The ontology provides the general pharmacological principles of the domain"[1].
1 Herrero-Zazo et al. http://ceur-ws.org/Vol-1114/Session3_Herrero-Zazo.pdf, 2013.
Drug Interaction Ontology (DINTO) 2/2
DINTO represents drug-drug interactions, not PDDIs. However, DINTO specifies a subclass of DDIs
named DDI described in a database. DDI in database represents "those DDIs imported
in DINTO from the DrugBank database with the purpose of distinguishing them from those inferred from the ontology”.
Yet, DrugBank contains PDDI information.
DIDEO: Methods (based on OBO Foundry principles)
Reuse of pre-existing ontologies Basic Formal Ontology
http://purl.obolibrary.org/obo/bfo.owl Drug Ontology
http://purl.obolibrary.org/obo/dron.owl Ontology of Biomedical Investigations
http://purl.obolibrary.org/obo/obi.owl Gene Ontology
http://purl.obolibrary.org/obo/go.owl Information Artifact Ontology
http://purl.obolibrary.org/obo/iao.owl Chemical Entities of Biological Interest
http://www.ebi.ac.uk/chebi/
Intended to be open source and community-driven
Four informational bases of PDDIs
1) Reasonable extrapolation 2) Physiological observations from clinical
studies 3) Drug-drug interaction observational data 4) Mechanistic assertions that are useful for
inferring drug-drug interactions
Four informational bases of PDDIs
Physiological experimentThe data establishes the
existence of a disposition borne by the enzyme.The drug metabolism
becomes part of an assay.
Substances come in portion!
Next steps Creating github project and an owl file for v1
of DIDEO (under development). coordinate with DINTO and DRON integrate what we learn from today's
workshop Iterative improvement of the representation
larger number of PDDI instances get integrated (next 12 month)
Coordinate with other stakeholders (NLM, FDA, NDF-RT, Cochrane Collaboration, W3C HCLS)
Acknowledgements For all authors: This project is supported by a grant from the National
Library of Medicine: “Addressing gaps in clinically useful evidence on drug-drug interactions” (R01LM011838-01) and the National Institute of Aging “Improving medication safety for nursing home residents prescribed psychotropic drugs” (K01 AG044433-01).
The authors thank Michel Dumontier and Alan Ruttenberg for their valuable comments, which have significantly improved the paper.
For PE: This work is supported by the National Center For Advancing Translational Sciences of the National Institutes of Health under Award Number KL2TR000146.
For DM: This work is partially supported by the Agency for Healthcare Research and Quality (AHRQ) Grant No. 1R13HS021826-01 (Malone DC-PI)
For JS: This work was carried out during the tenure of an ERCIM “Alain Bensoussan” Fellowship Programme. The research leading to these results has received funding from the European Union Seventh Framework Programme (FP7/2007-2013) under grant agreement no 246016.