<!DOCTYPE article PUBLIC "-//NLM//DTD JATS (Z39.96) Journal Archiving and Interchange DTD v1.0 20120330//EN" "JATS-archivearticle1.dtd">
<article xmlns:xlink="http://www.w3.org/1999/xlink">
  <front>
    <journal-meta />
    <article-meta>
      <title-group>
        <article-title>A Call for an Abductive­Reasoning Feature in  OWL­Reasoning Tools toward Ontology Quality Control</article-title>
      </title-group>
      <contrib-group>
        <aff id="aff0">
          <label>0</label>
          <institution>University of Colorado Denver, Department of Pharmacology</institution>
          ,
          <addr-line>MS 8303, RC­1 South, 12801 East 17</addr-line>
          ,
          <country country="US">USA</country>
        </aff>
      </contrib-group>
      <abstract>
        <p>   It is widely touted that OWL reasoners are able to improve the quality of ontologies  by making new inferences from and detecting inconsistencies among the assertions  that have been explicitly stated in the ontologies.  These reasoner actions are based on  standard deductive reasoning, which operates on the principle that assertions inferred  from premises that are assumed to be true also must be true.  Deductive reasoning is  similarly the basis for OBO­Edit, the primary ontology­management tool of the Open  Biomedical Ontologies (OBO), the most prominent and highly used set of ontologies  in the biomedical domain.      OBOs have typically been created modularly and with informal, natural­language  definitions; this is largely due to the fact that they are mostly developed by different  groups   (and   have   varying   levels   of   funding,   which   partly  accounts   for   the   varying  levels of quality among the OBOs).  There have been recent efforts to formalize and  link the  disparate  ontologies,  and there  is now an extensive  effort  within the OBO  Consortium to create formal definitions   of the component terms using more atomic  terms.   Running a deductive reasoner over these so­called cross­product definitions  has resulted in improved ontology quality, usually in the form of inferred is_a links  among   terms.     This   has   the   added   effect   of   aligning   the   linked   subject   and   object  terms; thus, the inferred  is_a  links point to what we refer to as  nonalignments, in  which either the subject terms are subsumptively linked and the object terms are not,  or vice versa.   We have additionally noticed that a form of abductive reasoning not  currently widely used has also proven useful in aligning linked terms and thus further  improving   ontology   quality.     First,   as   an   example   of   a   nonalignment   that   can   be  deductively detected:</p>
      </abstract>
      <kwd-group>
        <kwd>Class(positive regulation of hydrolase activity  complete</kwd>
        <kwd>  biological_process</kwd>
        <kwd>  restriction(regulates some hydrolase activity))</kwd>
      </kwd-group>
    </article-meta>
  </front>
  <body>
    <sec id="sec-1">
      <title>-</title>
      <p>an  is_a  link   is   inferred   between   the   defined   terms,   thus   pointing   to   the 
nonalignment:</p>
    </sec>
    <sec id="sec-2">
      <title>Class(coenzyme binding complete   molecular_function   restriction(results_in_binding_of some coenzyme))</title>
    </sec>
    <sec id="sec-3">
      <title>Class(FAD binding complete   molecular_function   restriction(results_in_binding_of some FAD))</title>
      <p>and the subsumption:</p>
    </sec>
    <sec id="sec-4">
      <title>SubClassOf(FAD coenzyme)</title>
      <p>   This is, of course, correct OWL behavior.  We have found in practice, however, 
that almost all occurrences of these types of nonalignments point to problems among 
the linked subject and object terms and that the subject and object terms should be 
aligned   in   the   large   majority   of   cases.     (That   is,   they   should   either   both   be 
subsumptively linked or both not subsumptively linked.)  In a study of nearly 8,000 
OBO cross­product definitions, 38.8% of nonalignments detected rely upon this type 
of   abductive   reasoning   (http://compbio.uchsc.edu/Hunter_lab/Bada/ 
nonalignments_2008_08_08.html).   Thus, we assert that a “button” for this type of 
reasoning   is   needed   in   ontology­management   tools,   analogous   to   the   classifying 
button that operates in tools such as Protege­OWL.  This type of abductive reasoning 
probably should not by default be enabled, but we assert this analysis of links and 
identified nonalignments would significantly improve ontology quality.</p>
    </sec>
  </body>
  <back>
    <ref-list />
  </back>
</article>