Successfully reported this slideshow.
We use your LinkedIn profile and activity data to personalize ads and to show you more relevant ads. You can change your ad preferences anytime.
Jodi Schneider & Richard D. Boyce
Department of Biomedical Informatics
School of Medicine, University of Pittsburgh
Design...
Upcoming SlideShare
Loading in …5
×

Acquiring and representing drug-drug interaction knowledge as claims and evidence, NLM informatics training conference, 2016-06-26

373 views

Published on

A poster for the National Library of Medicine Informatics Training Conference 2016: https://u.osu.edu/nlm2016/conference-agenda/

Potential drug-drug interactions (PDDIs) are a significant source of preventable drug-related harm. Poor quality evidence on PDDIs, combined with prescribers’ general lack of PDDI knowledge, results in thousands of preventable medication errors each year. One contributing factor is that PDDI knowledge lacks a standard computable format. To address this, we are researching efficient strategies for acquiring and representing PDDIs knowledge, focusing on assertions and their supporting evidence.

We are acquiring knowledge from several sources. First, we have transformed 410 assertions and 519 evidence items from prior work. Second, we are examining FDA-approved drug labels, and so far annotators have identified 609 evidence items relating to pharmacokinetic PDDIs from 27 FDA-approved drug labels. Third, annotators have found 230 assertions of drug-drug interactions in 158 non-regulatory documents, including full text research articles.

We are building a two-layer evidence representation, with both generic and domain-specific layers. The generic layer reuses the Micropublications Ontology to annotate assertions and their supporting data, methods, and materials. For the domain-specific component we are building DIDEO–the Drug-drug Interaction and Drug-drug Interaction Evidence Ontology. DIDEO adds specific knowledge, such as the study types required to establish a given type of PDDI. The current version of DIDEO has 385 subclass axioms, and reuses formalized knowledge items, including from the Drug Ontology, Chemical Entities of Biological Interest, the Ontology of Biomedical Investigations, and the Gene Ontology.

Published in: Technology
  • Be the first to comment

  • Be the first to like this

Acquiring and representing drug-drug interaction knowledge as claims and evidence, NLM informatics training conference, 2016-06-26

  1. 1. Jodi Schneider & Richard D. Boyce Department of Biomedical Informatics School of Medicine, University of Pittsburgh Designing    a  data  model  for  represen1ng  claims  and  evidence     1.  Iden1fy  key  ontologies  relevant  for  claims  and  evidence.   •  Nanopublica+ons  Ontology  represents  se4led  science:     Each  formalized  claim  (asser+on)  is  wrapped  in     provenance  and  publica+on  info.       •  Micropublica+ons  Ontology  represents  claims  and  evidence.  It  views  a  scien+fic  paper  as  a  network   of  claims  supported  by  data,  methods  and  materials.  Claims,  data,  methods,  and  materials  can  be   text,  images,  or  mul+media:  anything  the  Open  Annota+on  Ontology  can  reference.   2.  Iden1fy  key  domain  ontologies  to  reuse.   •  Ontology  of  Biomedical  Inves+ga+ons   •  Chemical  En++es  of  Biological  Interest   •  Drug  Ontology     3.  Conceptualize  3  layers  and  determine  what  belongs  in  each  layer:                   4.  Formalize  key  terms  about  drug-­‐drug  interac1ons  in  a  new  ontology,  DIDEO.   For  instance,  “poten+al  drug-­‐drug  interac+on”  gets  obo:DIDEO_00000000                         5.  Create  micropublica1ons  claims  and  nanopublica1on  asser1ons.   Micropublica+on  M:  “Clarithromycin  interacts  with  simvasta+n”   Nanopublica+on  asser+on  N:  obo:CHEBI_3732  obo:DIDEO_00000000  obo:  CHEBI_9150     6.  Connect  natural  language  quota1ons  and  their  informa1on-­‐retrieval  friendly  versions.   M  mp:formalizedBy  N  .   N  mp:formalizes  M  .           Ontologies,  Data,  and  Websites     DIDEO:  The  Poten+al  Drug-­‐drug  Interac+on  and  Poten+al  Drug-­‐drug   Interac+on  Evidence  Ontology   h4ps://github.com/DIDEO/DIDEO     Contribu+ons  to  Micropublica+ons  Ontology  (formalizes/formalizedBy):     h4ps://github.com/dbmi-­‐pi4/DIKB-­‐Micropublica+on/blob/master/data/ mp_1.18.owl     Drug  Interac+on  Knowledge  Base  website  &  discussion  forums   h4p://dikb.org    and  h4p://forums.dikb.org   Problem     Poten+al  drug-­‐drug  interac+ons  are  a  significant  source  of   preventable  drug-­‐related  harm.  The  drug  informa+on  sources   clinicians  use  are  disconcordant:  Most  drug  informa1on   sources  disagree  substan1ally  in  their  content  (e.g.  Abarca  et   al.  2003,  Wang  et  al.  2010,  Saverno  et  al.  2011).  This  problem   has  persisted  for  more  than  a  decade  (e.g.  Ayvaz  et  al.  2015,   Ekstein  et  al.  2015)  despite  extensive  editorial  work  on  the   part  of  each  drug  informa+on  source.  This  is  in  part  because:   (1)  There  is  no  standard,  agreed  upon  method  for  assessing   evidence  about  drug-­‐drug  interac+ons.   (2)  Knowledge  claims  and  evidence  about  drug-­‐drug   interac+ons  are  distributed  across  mul+ple  sources:     pre-­‐market  studies,  post-­‐market  studies,  and  clinical   experience.                           University  of  PiPsburgh   Department  of  Biomedical  Informa1cs  Funded  by  training  grant  T15LM007059    from  the  Na1onal  Library  of  Medicine/Na1onal  Ins1tute  of  Dental  and  Craniofacial  Research  and  by  R01LM011838    from  the  Na1onal  Library  of  Medicine.     Approach   Annota1on  Results       From  prior  work,  we  have  transformed   •  410  asser+ons   •  519  evidence  items     Annotators  have  also  iden+fied  evidence  in   •  158  non-­‐regulatory  documents     (including  full-­‐text  research  ar+cles)   •  27  FDA-­‐approved  drug  labels   This  leads  to  an  addi+onal:   •  230  asser+ons  of  drug-­‐drug  interac+ons  in  non-­‐regulatory  documents   •  609  evidence  items  rela+ng  to  poten+al  pharmacokine+c  drug-­‐drug   interac+ons  from  27  FDA-­‐approved  drug  labels   Acquiring  claims  and  evidence     1.   Formulate  claims  of  interest.   “Clarithromycin  interacts  with  simvasta+n”.     2.  Iden1fy  relevant  source  documents.   Source  documents  include  FDA-­‐approved  drug  product  labels  and  full-­‐text  research   papers  (clinical  trials  and  case  reports).     3.  Experts  assess  quality  &  relevance  of  source  documents.   Experts  check  that  documents  meet  inclusion  criteria.  Experts  find  relevant  claims,   methods,  and  results.       4.  Pre-­‐annota1on  by  computer  text  mining.   Source  documents  are  pre-­‐processed  to  find  drug  men+ons,  using  named  en+ty   recogni+on  algorithms.     5.  Human  curators  annotate  full-­‐text  documents.   (a)  The  curator  highlights  the  claim.   (b)  The  curator  enters  the  claim  and  scien+fic  method.   (c)  The  curator  is  prompted  to  add  data  based  on  the  method.     Claim 1 [Clarithryomycin interacts with Simvastatin] Data 1 Micropublication 1 mp:argues Method 1 mp:qualifies obo:CHEBI_3732 [Clarithryomycin] mp:qualifies obo:CHEBI_9150 [Simvastatin] mp:qualifies obo:DIDEO_00000000 [Potential drug-drug interaction] Materials 1 mp:supports mp:supports mp:supports mp:supports doi: "Clarithryomycin significantly (p<0.001) increased the AUC (and Cmax) of all 3 statins, most markedly simvastatin" oa:hasSource Publica1ons  and  Presenta1ons     1.  Jodi  Schneider,  Mathias  Brochhausen,  Samuel  Rosko,  Paolo  Ciccarese,  William  R.   Hogan,  Daniel  Malone,  Yifan  Ning,  Tim  Clark  and  Richard  D.  Boyce.  “Formalizing   knowledge  and  evidence  about  poten+al  drug-­‐drug  interac+ons.”  Interna7onal   Workshop  on  Biomedical  Data  Mining,  Modeling,  and  Seman7c  Integra7on  at   Interna7onal  Seman7c  Web  Conference  2015     h4p://ceur-­‐ws.org/Vol-­‐1428/BDM2I_2015_paper_10.pdf     2.  Jodi  Schneider,  Paolo  Ciccarese,  Tim  Clark  and  Richard  D.  Boyce.  “Using  the   Micropublica+ons  ontology  and  the  Open  Annota+on  Data  Model  to  represent   evidence  within  a  drug-­‐drug  interac+on  knowledge  base.”  4th  Workshop  on  Linked   Science  at  Interna7onal  Seman7c  Web  Conference  2014     h4p://ceur-­‐ws.org/Vol-­‐1282/lisc2014_submission_8.pdf     3.  Mathias  Brochhausen,  Jodi  Schneider,  Daniel  Malone,  Philip  E.  Empey,  William  R.   Hogan  and  Richard  D.  Boyce  “Towards  a  founda+onal  representa+on  of  poten+al   drug-­‐drug  interac+on  knowledge.”  First  Interna7onal  Workshop  on  Drug  Interac7on   Knowledge  Representa7on  at  the  Interna7onal  Conference  on  Biomedical  Ontologies   2014    h4p://ceur-­‐ws.org/Vol-­‐1309/paper2.pdf     4.  Mathias  Brochhausen,  Philip  E.  Empey,  Jodi  Schneider,  William  R.  Hogan,  and   Richard  D.  Boyce.  Adding  evidence  type  representa+on  to  DIDEO.  ICBO  2016     h4p://jodischneider.com/pubs/icbo2016.pdf     5.  Jodi  Schneider,  Samuel  Rosko,  Yifan  Ning,  and  Richard  D.  Boyce.  “Towards   structured  publishing  of  poten+al  drug-­‐drug  interac+on  knowledge  and  evidence.   Poster  presenta+on  at:  the  Pi4sburgh  Biomedical  Informa+cs  Training  Program  2015   Retreat.  Pi4sburgh,  PA,  August  20,  2015.     doi:10.6084/m9.figshare.1514991     6.  Jodi  Schneider  and  Richard  D.  Boyce  “Medica+on  safety  as  a  use  case  for   argumenta+on  mining”.  Dagstuhl  Seminar  16161:  Natural  Language  Argumenta+on:   Mining,  Processing,  and  Reasoning  over  Textual  Arguments,  Dagstuhl,  Germany,  April   19,  2016     h4p://www.slideshare.net/jodischneider/medica+on-­‐safety-­‐as-­‐a-­‐use-­‐case-­‐for-­‐ argumenta+on-­‐mining-­‐dagstuhl-­‐seminar-­‐16161-­‐2016-­‐0419       Future  Work   •  Build  an  informa+on  portal  that  supports  clinical  pharmacists  and  drug   informa+on  professionals  in  retrieving  the  claims  and  evidence.   •  Test  the  informa+on  portal  in  a  task-­‐based,  within-­‐subject,  user  study.   Measure  the  completeness  of  the  informa+on  experts  retrieve  with  our   informa+on  portal  compared  to  current  state-­‐of-­‐the-­‐art  retrieval  tools.   •  Test  the  feasibility  of    authors  annota+ng  their  own  claims  and  evidence.   •  Enable  annota+on  beyond  PubMed  Central  open  access  HTML.   •  Use  rulesets  of  “belief  criteria”  to  transform  evidence  to  a  knowledge  base.   Aim     In  this  work,  we  address  the  distributed  nature  of     drug-­‐drug  interac+on  knowledge,  by  developing  a    computable   representa+on  for  claims  and  evidence  about  drug-­‐drug   interac+ons.  Our  goal  is  to  support:   (a)  Knowledge  acquisi+on  from  full-­‐text  natural  language   (b)  Search  and  retrieval  of  all  evidence.   We  are  applying  this  representa+on  to  acquire  claims  and   evidence  about  pharmacokine+c  interac+ons  for  65  drugs.     This  will  help  us  design  a  search  portal,  to  test  whether   computable  representa+ons  of  knowledge  claims  and   evidence  can  improve  search  and  retrieval  of  poten+al  drug-­‐ drug  interac+ons.       Longer  term,  we  will  test  whether  this  can  help  reduce  the   disconcordance  between  different  drug  informa+on  sources.                 We  model  knowledge  as  claims   supported  by  evidence.                     asser+on   provenance   publica+on  info   nanopublica+on   DIDEO:  formalizing  medica1on  safety  studies     As  an  OWL  ontology,  DIDEO  supports  querying  as  well  as  reasoning.  For  instance,   curators  will  only  need  to  enter  a  few  facts  in  order  for  the  scien+fic  method  of  a   study  to  be  automa+cally  determined.     DIDEO  uses  ontological  realism  to  dis+nguish:   •  Poten+al  vs.  actual  drug-­‐drug  interac+ons.   •  Inferred  vs.  observed  interac+ons.   Method   Data   Claim   Dosage,  regimen,  clearance,  Cmax,  AUC,    half  life               Number  of  par+cipants,  randomiza+on   Claim,  data,  method,  and  material                                        

×