Building reference alignments for compound matching of multiple ontologies using OBO cross-products

Abstract

Existing ontology matching techniques are limited to matching two ontologies, but we argue that producing `compound' alignments, involving more than two ontologies, would be useful to support a next generation of semantic technologies. To foster the development of new techniques in this area, we have investigated the suitability of exploring OBO cross-products to derive ternary compound alignments that can be used as a benchmark. We were able to establish seven such reference alignments with over 100 mappings each, between ten biomedical ontologies. Preliminary experiments revealed that the increase in matching space and the inherently more di cult-to-compute ternary mapping pose interesting di culties to compound ontology matching.

-

Introduction. Both the `classical' and `complex' (e.g., [1{3]) ontology matching approaches focus on discovering mappings between two ontologies. We argue that it would be useful for the developers of ontology alignment systems to develop new techniques and tools for identifying 'compound matches', i.e. matches between class or property expressions involving more than two ontologies. The simplest of these mappings would correspond to an equivalence mapping between a class A of one ontology and an expression relating classes B and C of two other ontologies, constituting a ternary relationship. We investigate the suitability of exploring OBO cross-products to create ternary compound alignments between ontologies which can function as a gold-standard to support the evaluation of novel matching methods for compound alignment.

Approach. We consider that a ternary compound alignment is a set of correspondences (mappings) between classes from a source ontology Os and class expressions obtained by combining two other classes each belonging to a di erent target ontology Ot1 and Ot2. We de ne a ternary compound mapping as a tuple < X; Y; Z; R; M >, where X, Y and Z are classes from three distinct ontologies, R is a relation established between Y and Z to generate a class expression that is mapped to X via a mapping relation M. Some of the logical de nitions contained in OBO cross-products correspond to this type of mapping, for instance,

Pesquita et. al the class HP:0000337 labeled broad forehead is equivalent to an axiom obtained by relating the classes PATO:0000600 (increased width) and FMA:63864 (forehead ) via an intersection quali ed by an inheres in relation. We analyzed the resources available at obofoundry.org 1 and identi ed seven cross-products collections each with at least 100 de nitions corresponding to ternary compound mappings:

Source Ontology Target Ontologies Size

MP PATO UBERON 1725 HP PATO FMA 1519

MP PATO CL 407 WBPhenotype PATO GO 369

MP PATO GO 354 FYPO PATO GO 285

MP PATO NBO 100 To create the alignments based on the cross-products collections we used EDOAL [ 4 ], since it allows the construction of entities from other entities using algebraic operators. To represent intersection of we employed a class expression with the and operator.

Experiments. In ternary ontology matching, the search space is cubic, so matching even relatively small ontologies can pose e ciency problems. In a preliminary experiment, we adapted the anchor-based strategy of the AgreementMakerLight system[ 5 ] as well as its WordMatcher algorithm to use a modi ed Jaccard index that penalizes words shared by both target classes. We tested it in the MP-PATO-CL and MP-PATO-NBO alignments, obtaining recall values of 30 and 11% respectively, but precision values below 1%. These results highlight some of the complexity behind compound alignments, even between ontologies that strive to follow the same naming conventions. We posit that to solve these issues, background knowledge or instances would be needed to be able to discriminate between the candidate mappings.

Acknowledgements. This study was funded by the Portuguese FCT through the SOMER project (PTDC/EIA-EIA/119119/2010) and the LASIGE Strategic Project (PEst-OE/EEI/UI0408/2014) and by the National Science Foundation under award 1017225 \III:Small: TROn|Tractable Reasoning with Ontologies." 1 http://obofoundry.org/index.cgi?show=mappings

References

1. Ritze , D. , Volker, J., Meilicke , C. , Svab-Zamazal , O. : Linguistic analysis for complex ontology matching . Ontology Matching 1 ( 2010 )

2. Svab-Zamazal , O. , Svatek , V. , Iannone , L. : Pattern-based ontology transformation service exploiting oppl and owl-api. In: Knowledge Engineering and Management by the Masses . Springer ( 2010 ) 105 { 119

3. Meilicke , C. , Noessner , J. , Stuckenschmidt , H.: Towards joint inference for complex ontology matching . AAAI ( Late-Breaking Developments ) ( 2013 )

4. David , J. , Euzenat , J. , Schar e, F., Trojahn dos Santos, C. : The alignment api 4.0. Semantic web 2(1) ( 2011 ) 3 { 10

5. Faria , D. , Pesquita , C. , Santos , E. , Palmonari , M. , Cruz , I.F. , Couto , F.M.: The agreementmakerlight ontology matching system . In: On the Move to Meaningful Internet Systems: OTM 2013 Conferences , Springer ( 2013 ) 527 { 541