Reliable and interoperable computational molecular engineering: 2. Semantic interoperability based on the European Materials and Modelling Ontology
Martin Thomas Horsch, Silvia Chiacchiera, Youness Bami, Georg J. Schmitz, Gabriele Mogni, Gerhard Goldbeck, Emanuele Ghedini
aa r X i v : . [ c s . C E ] J a n Reliable and interoperable computationalmolecular engineering: 2. Semanticinteroperability based on the EuropeanMaterials and Modelling Ontology
Martin Thomas Horsch, Silvia Chiacchiera, Youness Bami, Georg J.Schmitz, Gabriele Mogni, Gerhard Goldbeck, and Emanuele Ghedini
Abstract
The European Materials and Modelling Ontology (EMMO) is atop-level ontology that was designed by the European Materials ModellingCouncil (EMMC) to facilitate semantic interoperability between platforms,models, and tools in computational molecular engineering, integrated com-putational materials engineering, and related applications of materials mod-elling and characterization. Additionally, domain ontologies exist based ondata technology developments from specific platforms. The present work dis-cusses the ongoing work on establishing a European Virtual MarketplaceFramework, into which diverse platforms can be integrated. It addresses com-mon challenges that arise when marketplace-level domain ontologies are com-bined with a top-level ontology like the EMMO by ontology alignment.
Semantic interoperability refers to an agreement between multiple software ordata infrastructures on the terms by which a scenario, i.e. , the circumstances
Martin Thomas Horsch and Silvia ChiacchieraUK Research and Innovation, STFC Daresbury Laboratory, Keckwick Ln, Daresbury,Cheshire WA4 4AD, UK, e-mail: { martin.horsch, silvia.chiacchiera } @stfc.ac.ukYouness Bami and Georg J. SchmitzACCESS e.V., Intzestr. 5, 52072 Aachen, Germany, e-mail: { y.bami,g.j.schmitz } @access.rwth-aachen.deGabriele Mogni and Gerhard GoldbeckGoldbeck Consulting Ltd, St John’s Innovation Centre, Cowley Rd, CambridgeCB4 0WS, UK, e-mail: { gabriele, gerhard } @goldbeck-consulting.comEmanuele GhediniUniversit`a di Bologna, Department of Industrial Engineering, Via Saragozza 8,40123, Bologna, Italy, e-mail: [email protected] 1 Horsch, Chiacchiera, Bami, Schmitz, Mogni, Goldbeck, Ghedini relevant to a particular application, can be described. In the presence of suchan agreement, different platforms can annotate data with metadata of thesame type and associated to each other by the same relations, so that theirmeaning is clarified and an exchange of information can rely on standard-ized semantics . This is independent of the precise technical mode in whichthe exchange is implemented, which additionally requires an agreement atthe syntactic level, e.g. , on the file formats or application programming in-terfaces that are employed for this purpose. There are common approachesby which semantic content can be serialized ( i.e. , denoted) straightforwardly, e.g. , using the JSON, YAML, TTL, and RDF/XML formats; obversely, se-mantic information can be included in file formats that permit the annotationof their content, such as CML [45, 47] and HDF5 [9, 38]. Accordingly, if thereis a common standard for the semantics, solutions for achieving syntacticinteroperability are immediately available as well. Semantic standards aredocumented by semantic assets , which include, in increasing order of expres-sivity [13],1. simple lists, where labels or names for concepts are collected without anyfurther elaboration,2. informal hierarchies, which include further information on the concepts,3. thesauri, where concept definitions are presented systematically in textform, possibly including certain relations, e.g. , between synonymous orantonymous concepts,4. taxonomies, i.e. , class hierarchies,5. conceptual models, e.g. , XSD or RDFS schemas, which extend the classhierarchy by a specification of possible contents or relations,6. ontologies, i.e. , class hierarchies + relation definitions + axioms.This is known as the hierarchy of semantic assets; levels 1 to 3 are targetedat human users, whereas at levels 4 to 6, a machine-processable formaliza-tion is usually given. This work deals with ontologies which, beside definingclasses and their properties in a way largely analogous to object-orientedprogramming, also include inference rules (or axioms) that can be used forautomated reasoning. The OWL language, for which a variety of formats ex-ist, is the established standard for formulating these definitions and rules;thereby, references to concepts from other ontologies or metadata schemascan be included, by which a semantic web is created in a similar way ashyperlinks are fundamental to the world wide web [1].The technical use of ontologies typically requires an agreement on data andmetadata related to one or multiple specific fields, or domains of knowledge,for which domain ontologies need to be developed in an effort requiring com-petencies from data management as well as concrete scientific or technical ex-pertise. In materials informatics, a variety of ontology-based approaches havebeen employed successfully to integrate data from diverse sources [31, 49].Moreover, domain-specific automated reasoning mechanisms can enhance the emantic interoperability based on the EMMO 3 functionality of cyber-physical and industry 4.0 infrastructures, supplement-ing the evaluation of discrete process models, such as Petri nets [46].Since ontologies can point to other ontologies, the semantic web exhibitsthe tendency for the semantic assets to aggregate into large connected net-works with a complex structure. If algorithmic logical reasoning is applied tothese structures, any inconsistencies become fatal, since a single contradictionat any point is sufficient to destroy the logical coherence of the entire contentas a whole. On the other hand, it is practically impossible to avoid mutual in-consistencies between any out of the multitude of domain ontologies that aredeveloped by different groups, coming from different fields, who may all havetheir own approach to the problem. This raises the challenge of establishingsemantic standards that apply at highest possible level of abstraction, underwhich all conceivable domain ontologies can be subsumed; these components,the design of which requires a solid philosophical underpinning, are known as top-level ontologies or upper ontologies [3, 8, 30]. The present work addressesthe problem of making a top-level ontology viable technically by connect-ing it to domain-specific ontologies and scenarios. The considered top-levelontology is the European Materials and Modelling Ontology (EMMO) [12],which is developed by the European Materials Modelling Council (EMMC)on the basis of previous efforts, cf. Schmitz et al. [39]. The relevant domainsof knowledge are related to applications of materials modelling, includingcomputational molecular engineering (CME), which has a focus on fluidsand interfacial phenomena [17, 20], and integrated computational materialsengineering (ICME), which has a focus on solids [38, 39].Knowledge bases that employ semantic technology can be split into twocomponents, one of which is terminological and deals with universals (here, classes ) and possible worlds, i.e. , what relations there can theoretically bebetween objects (here, individuals ); the second one is assertional and dealswith a factual scenario, i.e. , a description of the real world as it is, or witha concrete possible scenario that is not necessarily factual. Accordingly, theassertional component contains individuals and defines actual (rather thanmerely possible or necessary) relations between them. In description logic,these two components are referred to as the TBox and the ABox, and inmodel theory, the content of the ABox is referred to as a model [5]; here werefrain from using this term to avoid confusion with the concept of a modelin materials modelling in general and the EMMO in particular. Instead, theterm ontology will be used specifically for the terminological part (includingaxioms and rules), and scenario for a possible, concrete realization of the as-sertional part; this understanding of the term is in line with the philosophicalparadigm of nominalism, i.e. , non-existence of universals [6, 26, 36], and itsimplications on ontology design as implemented by the EMMO, which gener-ally avoids the explicit definition of individuals. This yields a strict separationbetween terminology (ontology), which applies to classes, and assertions thatapply to individuals. Accordingly, where an ontology definition file ( e.g. , inTTL format) includes individuals, as it is the case for the present domain
Horsch, Chiacchiera, Bami, Schmitz, Mogni, Goldbeck, Ghedini ontologies, cf.
Section 2, the latter are considered to be part of the scenario,not the ontology.The remainder of the present work is structured as follows: Section 2introduces ontologies used by a system of interoperable infrastructures forCME/ICME services: The European Virtual Marketplace Framework. Sec-tion 3 discusses the problem of matching top-level and domain ontologieswith each other, from a general point of view, and Section 4 applies this tothe challenge of describing concrete features of services provided at the Eu-ropean Virtual Marketplace Framework in agreement with the joint top-levelinteroperability standard which is given by the EMMO; to illustrate this, ascenario from molecular modelling is considered in detail. Conclusions areformulated in Section 5. Work in progress on representing EMMO scenariosas graphs and using these graphs to automatically create Python objects isdocumented in the Appendix.
The European Virtual Marketplace Framework, established by the joint workof the MarketPlace and Virtual Materials Marketplace (VIMMP) consortiain coordination with the EMMC, is open to participation by any interestedprovider, translator ( i.e. , facilitator), or end user of CME/ICME services.It is entirely based on transparent and openly accessible specifications, inparticular, on a coherent system of ontologies with the EMMO at the toplevel. By creating an open framework on the basis of community-governedinteroperability standards, a variety of projects, many of which (includingEMMC-CSA, MarketPlace, VIMMP, and SimDOME) are funded from theLEIT-NMBP line of the European Union’s Horizon 2020 research and inno-vation programme, contribute to a system of platforms and infrastructuresthat will support the uptake of CME/ICME solutions by industrial researchand development practice.The European Virtual Marketplace Ontology (EVMPO) was agreed as acommon point of departure for the standardization of service-oriented seman-tics, relevant to marketplace platforms, by the projects involved in establish-ing the European Virtual Marketplace Framework. By defining eleven funda-mental paradigmatic categories, which correspond to irreducible terms thatare constitutive to the paradigm underlying materials modelling marketplaces ,the EVMPO provides a structure for further, more specific marketplace-levelontologies. These fundamental paradigmatic categories are specified in Sec-tion 2.2. emantic interoperability based on the EMMO 5
Terms which are not closely related to the marketplace paradigm itself,but occur in the related semantic web, are defined to be non-paradigmatic .For this purpose, the EVMPO includes evmpo:annotation as a fundamen-tal non-paradigmatic category. Below the fundamental level, the EVMPOalso includes non-fundamental entites as subclasses, e.g. , evmpo:simulation as a subclass of evmpo:process , and evmpo:service as a subclass of evmpo:product . Fig. 1 shows how, in terms of the rdfs:subClassOf relation,this class hierarchy can be integrated with the most closely related entitiesfrom the EMMO. It should be noted that the EMMO, and consequently alsothe integration as outlined in Fig. 1, is work in progress and may thereforebe subject to significant extensions and modifications in the future.Consistency with the EVMPO, and by implication consistency with theEMMO (subsequent to a release of the first stable version of the EMMO [12]),is a requirement for all components and infrastructures that aim at interop-erating within the European Virtual Marketplace Framework. This designfacilitates that VIMMP, MarketPlace, and others can agree on the definitionof paradigmatic entities as well as annotations which are either directly orindirectly related to paradigmatic entities, while any platform retains theoption to extend its own semantic base as required. To remain interoperablewithin the European Virtual Marketplace Framework, any such additionalworks need to be connected semantically to the EVMPO. In this way, amulti-tier system of ontologies is established as follows, from the top (mostgeneral) to the lowest (most specific) level: • Top-level ontology: The top level is specified within the EMMO [12]. • Marketplace-level fundamentals: The fundamental categories and theirmost significant subclasses are specified within the EVMPO. • Marketplace-level ontologies: A more detailed development of the EVMPOcategories, with a focus on service and model interoperability, is coveredby domain ontologies at the marketplace level [19]. The structure of thissystem of ontologies mostly follows the subdivision of the semantic spaceinto fundamental paradigmatic categories as specified by the EVMPO,cf. Section 2. • Subdomain-specific ontologies: These ontologies deal with subcategories; e.g. , subdividing the semantic space covered by the VIMMP Software On-tology (VISO), specific modelling and simulation granularity levels are ad-dressed by VISO-EL (electronic), VISO-AM (atomistic and mesoscopic),and VISO-CO (continuum) [20].The design of the marketplace-level ontologies is guided by two principles:First, they should include classes and relations that are suitable to de-scribe services, agents, and other aspects that are relevant to infrastructureswhich are expected to interoperate within the European Virtual MarketplaceFramework [19]. Beside materials modelling marketplaces, this prospectivelyincludes data marketplaces, innovation platforms, translation environmentsand simulation platforms as well as any resources connected to such infras-
Horsch, Chiacchiera, Bami, Schmitz, Mogni, Goldbeck, Ghedini (cid:1)(cid:2)(cid:3)(cid:4)(cid:5)(cid:6)(cid:7)(cid:8)(cid:9)(cid:1)(cid:10)(cid:11)(cid:6)(cid:12)(cid:4)(cid:4) (cid:1)(cid:7)(cid:10)(cid:7)(cid:13)(cid:5)(cid:14)(cid:15)(cid:7)(cid:16)(cid:5)(cid:6)(cid:9)(cid:12)(cid:17)(cid:16)(cid:5)(cid:16)(cid:3)(cid:6)(cid:7)(cid:8)(cid:12)(cid:17)(cid:13)(cid:7)(cid:10)(cid:9)(cid:12)(cid:18)(cid:12)(cid:17)(cid:16)(cid:5)(cid:17)(cid:16)(cid:12)(cid:10)(cid:1)(cid:10)(cid:12)(cid:16)(cid:12)(cid:10) (cid:1)(cid:10)(cid:11)(cid:1)(cid:12)(cid:10)(cid:16)(cid:3)(cid:7)(cid:4)(cid:4)(cid:12)(cid:4)(cid:4)(cid:15)(cid:12)(cid:17)(cid:16)(cid:6)(cid:11)(cid:15)(cid:15)(cid:19)(cid:17)(cid:5)(cid:6)(cid:7)(cid:16)(cid:5)(cid:11)(cid:17)(cid:15)(cid:11)(cid:13)(cid:12)(cid:8) (cid:5)(cid:17)(cid:20)(cid:11)(cid:10)(cid:15)(cid:7)(cid:16)(cid:5)(cid:11)(cid:17)(cid:9)(cid:6)(cid:11)(cid:17)(cid:16)(cid:12)(cid:17)(cid:16)(cid:9)(cid:12)(cid:17)(cid:16)(cid:5)(cid:16)(cid:3)(cid:5)(cid:17)(cid:20)(cid:10)(cid:7)(cid:4)(cid:16)(cid:10)(cid:19)(cid:6)(cid:16)(cid:19)(cid:10)(cid:12)(cid:1)(cid:10)(cid:11)(cid:13)(cid:19)(cid:6)(cid:16)(cid:14)(cid:11)(cid:11)(cid:13)(cid:4)(cid:12)(cid:10)(cid:18)(cid:5)(cid:6)(cid:12)(cid:15)(cid:7)(cid:16)(cid:12)(cid:10)(cid:5)(cid:7)(cid:8)(cid:4)(cid:5)(cid:15)(cid:19)(cid:8)(cid:7)(cid:16)(cid:5)(cid:11)(cid:17) (cid:1)(cid:10)(cid:11)(cid:6)(cid:12)(cid:4)(cid:4)(cid:12)(cid:15)(cid:15)(cid:11)(cid:12)(cid:17)(cid:14)(cid:5)(cid:17)(cid:12)(cid:12)(cid:10)(cid:12)(cid:13)(cid:9)(cid:12)(cid:17)(cid:16)(cid:5)(cid:16)(cid:3)(cid:1)(cid:2)(cid:3)(cid:4)(cid:5)(cid:6)(cid:7)(cid:8)(cid:1)(cid:10)(cid:11)(cid:6)(cid:12)(cid:4)(cid:4) (cid:4)(cid:12)(cid:15)(cid:5)(cid:11)(cid:4)(cid:5)(cid:4)(cid:15)(cid:7)(cid:16)(cid:12)(cid:10)(cid:5)(cid:7)(cid:8)(cid:12)(cid:21)(cid:5)(cid:4)(cid:16)(cid:12)(cid:17)(cid:16)(cid:5)(cid:17)(cid:16)(cid:12)(cid:10)(cid:1)(cid:10)(cid:12)(cid:16)(cid:12)(cid:10)(cid:4)(cid:5)(cid:14)(cid:17)(cid:4)(cid:12)(cid:15)(cid:5)(cid:11)(cid:16)(cid:5)(cid:6) (cid:1)(cid:10)(cid:11)(cid:1)(cid:12)(cid:10)(cid:16)(cid:3)(cid:15)(cid:11)(cid:13)(cid:12)(cid:8)(cid:4)(cid:3)(cid:15)(cid:22)(cid:11)(cid:8)(cid:5)(cid:6)
Fig. 1
Fundamental paradigmatic categories and other classes from the EVMPO(ellipses) together with related classes from the EMMO (rectangles); arrows betweenclasses represent the transitive reduction of the rdfs:subClassOf relation, and doublelines between classes represent the owl:equivalentClass relation. Note: The EVMPOand the EMMO are work in progress; this diagram refers to EMMO pre-release version0.9.10 [12].emantic interoperability based on the EMMO 7 tructures (e.g., external databases) and services associated with them. As asecond principle, the structure and scope of the individual marketplace-levelontologies should follow the subdivision of the semantic space into fundamen-tal paradigmatic categories from the EVMPO; by implication, any entitiesthat cannot be subsumed under any of these categories need to be classifiedas subclasses of evmpo:annotation .The alignment of the eight marketplace-level ontologies MACRO, MMTO,OSMO, OTRAS, VICO, VISO, VIVO, and VOV [19] with the fundamentalparadigmatic categories from the EVMPO is specified in Section 2.2. Section2.3 summarizes how models and the associated variables can be described inthis context; it introduces the example scenario that will be used in Section 4to illustrate a possible way of matching terms from the marketplace-levelontologies with terms from the EMMO.
Below we list the fundamental paradigmatic categories from the EVMPOand their definitions, including a description of selected subclasses of thefundamental categories and the related marketplace-level ontologies [19] fromthe VIMMP project:1. evmpo:assessment , i.e. , a proposition on the accuracy or performance ofan entity or an a expression of trust in an entity −→ evmpo:endorsement assessment , i.e. , an assessment by which anentity is claimed to be good or to be fit for a certain purpose −→ evmpo:requirement assessment , i.e. , an assessment concerningcomputational requirements ( e.g. , computing time, memory, or hardwareand software prerequisites) −→ evmpo:validity assessment , i.e. , an assessment concerning theuncertainty or error associated with a data item or with a model or simu-lation workflow by which data are generated Marketplace-level ontology:
VIMMP Validation Ontology (VIVO)2. evmpo:calendar event , i.e. , a meeting or activity which is scheduledor can be scheduled; this is defined to be equivalent with the entity Vevent from the W3C iCalendar ontology with time zones as datatypes(ICALTZD) [11]
Marketplace-level ontology:
Ontology for Training Services (OTRAS)3. evmpo:communication , i.e. , any message that is communicated −→ evmpo:declaration , i.e. , a communication without a recipient −→ evmpo:interlocution , i.e. , a communication with recipient(s) −→ evmpo:statement , i.e. , an elementary communication that cannotbe decomposed into multiple statements Marketplace-level ontology:
VIMMP Communication Ontology (VICO)
Horsch, Chiacchiera, Bami, Schmitz, Mogni, Goldbeck, Ghedini evmpo:information content entity , e.g. , a journal article or a graph;this is defined to be equivalent with IAO 0000030 , labelled informationcontent entity , from the Information Artifact Ontology (IAO) [10]
Marketplace-level ontology:
Ontology for Training Services (OTRAS)5. evmpo:infrastructure , i.e. , infrastructure of an EVMPO interoperableplatform ( e.g. , related to data, hardware, and software) Marketplace-level ontologies:
Marketplace-Accessible Computational Re-source Ontology (MACRO), VIMMP Software Ontology (VISO)6. evmpo:interpreter , corresponding to the concept from Peirce’s semiotics,which is based on the triad sign, interpreter, object ; this is defined to beequivalent with emmo-semiotics:interpreter −→ evmpo:agent , i.e. , an interpreter that can interact with an infras-tructure ( e.g. , with a virtual marketplace platform) Marketplace-level ontology:
VIMMP Communication Ontology (VICO)7. evmpo:material , i.e. , an amount of a physical substance (or mixture ofsubstances) that is part of a more comprehensive real-world object [12];this is defined to be equivalent with emmo-material:material evmpo:model , i.e. , an entity that represents a physical object or processby direct similitude or by capturing the relations between its proper-ties in a mathematical framework; this is defined to be equivalent with emmo-models:model Marketplace-level ontologies:
Materials Modelling Translation Ontology(MMTO), Ontology for Simulation, Modelling, and Optimization (OSMO),VIMMP Software Ontology (VISO)9. evmpo:process , i.e. , the temporal evolution of one or multiple entities −→ evmpo:business process , i.e. , an abstract procedural representa-tion of economic relations −→ evmpo:physical process ; this is defined to be equivalent with emmo-processual:process Marketplace-level ontology:
Ontology for Simulation, Modelling, and Op-timization (OSMO)10. evmpo:product , i.e. , a good or service (which can be offered either on avirtual marketplace or off-site) Marketplace-level ontologies:
Materials Modelling Translation Ontology(MMTO), Ontology for Training Services (OTRAS)11. evmpo:property , i.e. , an entity that is determined by an observation pro-cess, involving a specific observer that perceives or measures it; this isdefined to be equivalent with emmo-properties:property Marketplace-level ontology:
VIMMP Ontology of Variables (VOV)The fundamental paradigmatic categories need not be disjoint; e.g. , evmpo:ma - terial and evmpo:product overlap, since a material can be manufacturedwith the intent of selling it as a commodity, by which it becomes a product.Concerning category 9, n.b. , due to including business processes, whichare not necessarily processes in the sense of the EMMO, evmpo:process isa proper superclass of emmo-process:process . To facilitate ontology align- emantic interoperability based on the EMMO 9 ment, the subclass evmpo:physical process was introduced, which is equiv-alent to emmo-process:process . Subclasses of evmpo:physical process in-clude evmpo:manufacturing process , i.e. , a physical process that serves theproduction of a good, and evmpo:simulation , i.e. , a simulation workflow. The purpose of the VIMMP Ontology of Variables (VOV) consists in orga-nizing the variables (in a broad sense, including constants) that appear inmodelling and simulation, and to connect them to models and algorithms inwhich they are involved as well as to model-related entities ( e.g. , interactionsites) to which they are attached. In VISO, these model-related entities arereferred to as model objects (class viso:model object ).In particular, the marketplace-level ontologies VOV, VISO, and OSMOcan be combined to characterize models, algorithms, and workflows. Two re-lations from VOV that serve this purpose are vov:involves and vov:has at - tached . A triple of the type :X vov:involves :Y describes that there is amathematical expression, extension, or algorithmic formulation of :X thatcontains the variable or the model object :Y ; therein, :X can be a model, aconstituent part of a model ( e.g. , a physical equation or materials relation),a solver (numerical implementation of a model), or a physical entity thatis modelled. A triple of the type :X vov:has attached :Y is employed toassociate a model object :X with a variable :Y . In this way, VOV, VISO, andOSMO can be used to state that a rigid two-site Mie potential is employedas model a for ammonia: molmod:AMMONIA a osmo:einecs_listed_material;osmo:has_ec_number "231-635-3"^^xs:string. The scenario given by the statements above will be employed below as anexample for illustrating the procedure of matching marketplace-level ontolo-gies with the EMMO. The problem in general and the present procedure areintroduced in Section 3. In Section 4.1, the present scenario is reexpressed interms of classes and relations from the EMMO, which inevitably results in a loss of information; in Section 4.2, conceivable (“candidate”) correspondencesresulting from this transposition are evaluated and reformulated, contributingto an ontology alignment.Further triples can provide more information, e.g. , on the size parameterof the Mie interaction sites, which is here specified to be σ = 2 . molmod:NH3_POTENTIALosmo:has_aspect_object_content molmod:NH3_POTENTIAL_SITE_A.molmod:NH3_POTENTIAL_SITE_A a osmo:materials_relation;osmo:has_aspect_object_content molmod:NH3_CONDITION_POT_A;vov:involves molmod:NH3_SITE_A.molmod:NH3_CONDITION_POT_A a osmo:quantitative_condition;osmo:contains_predetermined_variable molmod:NH3_PARAMETER_SIG.molmod:NH3_PARAMETER_SIG a osmo:unique_elementary;osmo:has_variable_name "sigma"^^xs:string;osmo:has_initial_elementary_value molmod:NH3_ELEMENTARY_SIG;osmo:has_variable_unit qudt-unit:Angstrom.molmod:NH3_ELEMENTARY_SIG a osmo:elementary_value;osmo:is_decimal 2.5764. Therein, the entity representing the unit ˚A is obtained from the system ofontologies for quantities, units, and datatypes (QUDT) [48].
A major design goal for a top-level ontology (see Mascardi et al. [29] foran illustrating comparison of popular ones) consists in achieving the de-sired level of expressivity with a minimal repertoire of basic terms and re-lations. Obversely, to ensure interoperability at the marketplace level, theemployed ontologies need to capture detailed characteristics of many par-ticular services, models, and interactions. Accordingly, the structure of thecorresponding semantic space at the lower level is comparably complex; e.g. , the marketplace-level ontologies from VIMMP contain over 400 ob-ject class definitions (and, for technical reasons, over 800 additional defi-nitions of property classes), over 250 definitions of object properties ( i.e. ,instances of owl:ObjectProperty ), and over 100 definitions of data prop-erties ( i.e. , instances of owl:DatatypeProperty ). Therefore, by design, theEMMO needs to have a structure that is substantially different from thatof the marketplace-level ontologies [19]. To ensure that the EMMO is consis-tently employed at all levels, so that it can contribute to platform and service emantic interoperability based on the EMMO 11 interoperability as far as possible, the marketplace-level ontologies need tobe aligned with the EMMO. Before returning to this specific problem, thepresent section summarizes some of the related theoretical concepts.In principle, semantic assets are designed to allow data integration andovercome the data heterogeneity problem; in reality, semantic heterogeneitydoes arise, and it grows over time as resources are added to the semantic web.This is known as the Tower of Babel problem [21, 22]. While some authorsregard any presence of semantic heterogeneity as a failure of semantic inter-operability and hope for universal agreements, others think that it is unavoid-able and look for strategies to deal with it. This may involve a standardizedway of documenting semantic assets, basic agreements on the approach toontology design ( e.g. , formal monism), and meta-ontological formalizationsof roles, procedures, and good practices (or best practices), aiming at prag-matic interoperability [4, 18, 37, 40]. For this approach, the challenge consistsin agreeing and specifying how the semantic space is structured, documented,and employed in practice; by raising the domain for which universal agree-ments are pursued from the ontological level to the meta-ontological level,“the Tower of Babel becomes a Meta-Tower of Babel” [43].As a consequence, semantic heterogeneity is seen as a necessary propertyof the semantic web, and ontology matching and integration become basicfeatures of its successful mode of operation, rather than an expression ofincompleteness. Options for implementing such a mode of operation havebeen extensively discussed in the literature, first for schemas and then forontologies; for example, cf.
Noy [34], Euzenat and Shvaiko [16], the OntologyMatching website [42], and the material from a series of events organizedby the Ontology Alignment Evaluation Initiative (OAEI) [23]. The commonchallenge is how to make use of the knowledge represented in two ontolo-gies, which can differ at various levels (language used, expressivity, modellingparadigm, etc. ). Typically, such challenges arise if there is an overlap in thedomains of knowledge addressed by multiple ontologies, such that data an-notated in diverse ways need to be combined and processed together, or if aplatform employs multiple domain ontologies that are based on different top-level ontologies. Typical applications include, e.g. , simultaneous querying ofmultiple knowledge bases or, as addressed here, the transposition of semanticcontent from a source ontology s to a target ontology t . Requirements generally differ between tasks occurring at runtime (emphasison efficiency) and at design time (emphasis on completeness). Similarly, thedesired outcome is not always the same: As opposed to ontology integration, i.e. , the creation of a merged (integrated) ontology, ontology matching is aprocess that yields a set of correspondences: An ontology alignment [16, 41].
This is known as the ontology matching problem (OMP), for which a greatvariety of strategies and algorithms have been devised [2, 16, 28, 31, 41].The correspondences can be formulated, with increasing expressive power, asfollows:1. As tuples ( σ, τ, ω ), where σ and τ are classes or relations from the sourceand target ontologies s and t , respectively, and ω ∈ {⊑ , ⊒ , ≡} is a logicaloperator indicating whether σ is a subclass or subproperty ( i.e. , σ ⊑ τ )or a superclass or superproperty of τ ( i.e. , σ ⊒ τ ), or whether the two areequivalent ( i.e. , σ ≡ τ ). If a statistical approach is employed, the tuplemay contain an additional metadata item indicating the probability thatthe operator can be applied; cf. Shvaiko and Euzenat [41] for a discussionof such approaches.2. As OWL statements ( e.g. , in TTL format) or equivalently as formulæ inOWL description logic (OWL DL); where viable, this approach is prefer-able in the present context, since it can be implemented immediately in anytechnical infrastructure where the marketplace-level ontologies are used.3. In more expressive logics, e.g. , first-order logic or description logics thatextend OWL DL [5], or by rewriting rules constituting, e.g. , a graph trans-formation system [24]. While these formalisms are undecidable in general,their particular application to the OMP might restrict formulæ and rulesin a way that keeps them decidable and tractable computationally; thiswill be addressed by future work, since as discussed below, OWL DL isinsufficient to capture a number of typical and relevant cases.For the present purpose, we will assume that the source ontology, whichhere is a marketplace-level ontology, is more expressive semantically, and lessabstract; the EMMO will be the target ontology.Since ontologies can easily involve hundreds of entities, automatic andsemi-automatic tools with different levels of user involvement have been cre-ated to perform ontology matching; see, e.g. , Tab. 2 from Mart´ınez and Al-dana [28] for a list of algorithms, and Tab. 1 from Shvaiko and Euzenat [41] fora list of tools. The strategy they use to find correspondences is to look for sim-ilarities between entities (classes, relations, individuals), considering variousaspects: Label and documentation, using exact or approximate text match-ing (terminological/linguistic aspect); subclasses and superclasses, relations,their domain and ranges (structural aspect); instances of classes (extensionalaspect); data type (constraint-based aspect); class descriptions in the senseof rules (semantic aspect) [34, 41]. Different matchers can be combined, withweights and thresholds, to build a meta-matching approach [28]; an API todescribe alignments has been proposed as well [14].To be able to compare possible candidate correspondences, distances andsimilarities measures can be defined [28]. The quality of the overall alignmentcan also be evaluated, using precision (a measure of correctness) and recall (ameasure of completeness); a combination of the two is called F-measure [41].These concepts were borrowed from the area of data retrieval; in the context emantic interoperability based on the EMMO 13 of ontology alignment, they can be defined as follows: “given a referencealignment R , the precision of an alignment A is P ( A, R ) = | R ∩ A | / | A | , andthe recall is given by R ( A, R ) = | R ∩ A | / | R | ” (Euzenat [15], Definition 9).Note that for this purpose of evaluation and comparison of alignments, areference alignment, for example compiled by a human, has to be provided.To support the matching process, external resources can be used, e.g. , ahigher-level ontology operating at a moderate level of abstraction [30] or alinguistic resource such as WordNet [32].Applying the OMP to the particular problem at hand, this work aims atestablishing a limited semantic space which is aligned between the two levels,serving both as a proof of concept and as a basis for future work that willfurther extend the aligned space. In the present case, the concerned ontolo-gies are all in an early stage of development. As a consequence, substantialcorpora of triples formulated in any of the ontologies do not exist; any at-tempt at matching the marketplace-level ontologies with the EMMO needsto be based on limited sets of triples that were either created for this spe-cial purpose or to guide the marketplace platform design. This precludes anyautomated approach to ontology matching. Instead, the present strategy ispurely based on (human-)asserted correspondences of classes and relations.It can be summarized as follows; we assume that in the beginning, a descrip-tion in terms of the source ontology s is available for a scenario that is wellunderstood by the ontologist who carries out the matching:1. Reexpress the scenario in terms of the target ontology t ; observe howsource-ontology terms σ ◦ i , i.e. , classes, relations, or more complex expres-sions, are mapped to target-ontology terms τ ◦ i , where i ∈ N is an index.These instances, σ ◦ i τ ◦ i , yield candidate correspondences σ ◦ i ⊑ ? τ ◦ i , eachof which is further addressed by the subsequent steps.2. If the candidate correspondence is valid, σ ◦ i ⊑ τ ◦ i , proceed to the next stepwithout any change, i.e. , σ ′ i = σ ◦ i , τ ′ i = τ ◦ i . If it is invalid, σ ◦ i τ ◦ i , relaxit such that σ ′ i ⊑ τ ′ i becomes valid by • replacing τ ◦ i with a more general term τ ′ i ⊒ τ ◦ i ( τ -generalization), or • replacing σ ◦ i with a more specific term σ ′ i ⊑ σ ◦ i ( σ -refinement).If this attempt reduces the correspondence to a trivial statement, discardthe candidate correspondence.3. Strengthen the correspondence as far as possible by • replacing σ ′ i with a more general term σ ′′ i ⊒ σ ′ i ( σ -generalization), • replacing τ ′ i with a more specific term τ ′′ i ⊑ τ ′ i ( τ -refinement), or • replacing the subclass/subproperty operator ⊑ with the equivalenceoperator ≡ ( σ - τ -identification).The outcome has the form σ ′′ i ⊑ τ ′′ i , or σ ′′ i ≡ τ ′′ i if the operator has beenreplaced.
4. Express the correspondence in OWL ( e.g. , TTL notation) or OWL DL ifpossible. If it cannot be expressed in OWL, attempt to simplify it, e.g. , byrelaxing it again as far as necessary; if that is impossible without reducingthe correspondence to a trivial statement, the candidate is discarded.Step 4 can easily be adjusted in cases where there is no need to introduce thealignment immediately into an OWL based architecture.This procedure is not meant to be automated; in particular, for the presentproblem of matching domain ontologies with the EMMO pre-release version,it requires a close reading of definitions from the source and target ontologiesand the available documentation to assess how the statements must be re-laxed and strengthened, respectively, to be valid and as powerful as possible.
The following remarks refer to the EMMO pre-release development version0.9.10 [12]; the precise way of mapping individuals and relations betweenthem from the marketplace-level ontology representation to an EMMO rep-resentation will be subject to change as all of the involved ontologies arefurther developed. However, the main purpose of the present example con-sists not in its result, but in taking the first steps toward ontology alignmentwithin the European Virtual Marketplace Framework, evaluating to whatextent the method proposed in Section 3.2 is suitable for this purpose, andproviding an orientation for future work on this basis.Mapping the first part of the example from Section 2.3 to entities fromEMMO version 0.9.10 yields: molmod:AMMONIA a emmo-material:material.molmod:NH3_POTENTIAL a emmo-models:material_relation.molmod:AMMONIA emmo-models:has_model [a emmo-models:physics_based_model;emmo-mereotopology:has_proper_part molmod:NH3_POTENTIAL,molmod:NH3_RIGID_UNIT].molmod:NH3_RIGID_UNIT a emmo-graphical:symbolic, emmo-semiotics:sign;emmo-mereotopology:has_proper_part molmod:NH3_SITE_A,molmod:NH3_SITE_B, molmod:NH3_SITE_COM.molmod:NH3_SITE_A a emmo-graphical:symbol.molmod:NH3_SITE_B a emmo-graphical:symbol.molmod:NH3_SITE_COM a emmo-graphical:symbol. emantic interoperability based on the EMMO 15
To keep the example above and the present discussion readable, EMMOentities are represented by their labels rather than their identifiers; e.g. , wewrite emmo-models:has model to represent the relation that has the identifier emmo-models:EMMO 24c71baf 6db6 48b9 86c8 8c70cf36db0c and the label "has model" .The transposed version above does not include any entity correspondingto the elementary data item from the source-ontology version of the example, i.e. , the string "231-635-3" , since the present target ontology does not in-clude any data properties and elementary data items; in the EMMO, strings,numbers, etc. , are emmo-graphical:symbolic individuals rather than ele-mentary data items, and all relations that are defined in the EMMO areobject properties. Accordingly, it would be possible to introduce an EMMOindividual representing "231-635-3" ; however, since it is impossible in OWLfor an elementary data item to be the same as ( owl:sameAs ) an individual,this would not yield any formal correspondence at the ontological level. Thesame, consequently, applies to the datatype properties from the marketplace-level ontologies: They cannot be matched with any of the relations from theEMMO; the EMMO only contains object properties, which cannot be equiv-alent ( owl:equivalentProperty ) to datatype properties.Considering the present example, it should be noted that entities suchas the pair potential or the molecular centre of mass from the example arenot actual features of fluid ammonia, but of a molecular model for ammo-nia. For the transposition of the scenario, we assume that in an appropriateformal language for molecular models, interaction sites and other structure-less objects ( viso-am:structureless object ) are syntactically irreducibleelementary symbols and hence emmo-graphical:symbol individuals by theEMMO definition [12], whereas a rigid unit that contains multiple sites is anentity composed of multiple symbols (in the theory of formal languages, a word ) and hence a emmo-graphical:symbolic entity, which in the EMMOis defined as a composition of symbols, including a single symbol as a specialcase [12]. The specification of such a formal language, which would be un-problematic, is beyond the scope of the present work. We merely note thatby the EMMO definition, symbols “of a formal language” are the smallestirreducible parts of symbolic entities, which depends on the way in whichthe formal language is specified; it is certainly possible to refer to interactionsites and other structureless objects by elementary symbols ( a , b , c , . . . , orspecial symbols or elementary keywords devised for this purpose).By its foundation on Peircean semiotics [35], the EMMO applies a cleardistinction between physical objects and the signs for them; thereby, symbolicentities (including irreducible symbols) need not be signs; they only becomesigns if they represent existing physical objects in the eyes of an interpreter.(Here, the interpreter, who is not explicitly represented in the example, mightbe a provider or user of the molecular model.) Moreover, due to the interpreta-tion of mereotopology [44] underlying the EMMO, any existing (“real world”)physical objects must be four-dimensional spacetime entities; i.e. , they must correspond to trajectories of contiguous volumes that can be tracked overtime. Accordingly, the definition of the top entity from EMMO version 0.9.10, emmo-mereotopology:emmo , states: “A real world object is then a 4D topo-logical sub-region of the universe. [. . . ] It follows that, for the EMMO, realworld objects of dimensionality lower than 4D do not exist ( e.g. surfaces,lines)” [12]. If a symbolic entity is used to refer to any lower-dimensionalstructure, it is not a emmo-semiotics:sign (nor an emmo-models:model ),since it does not represent a “real-world object” in the sense attributed tothis term by the EMMO.Hence, in particular, molmod:NH3 RIGID UNIT , the rigid unit that con-tains the interaction sites of the molecular model, is a sign for an ammoniamolecule. It is debatable whether it is also a model in the sense of the EMMO,which requires “direct similitude” with the represented object or a mathe-matical formalization of its behaviour [12]; here, we do not classify it as amodel. The two Mie sites molmod:NH3 SITE A and molmod:NH3 SITE B , how-ever, are only symbols (and part of a sign), since they do not represent anyparticular part of the molecule or any other part of spacetime specifically. Thepair potential molmod:NH3 POTENTIAL might be understood to represent thepotential energy from the pairwise interaction between two NH molecules;however, this does not constitute semiosis in the sense of the EMMO, sincethe referenced object is not a 4D entity, but a hypothetical additive con-tribution to a scalar quantity. Consequently, molmod:NH3 POTENTIAL is not an emmo-models:model individual; however, it is a part of such an individ-ual, since the combination of all pair potentials involved in modelling a sys-tem yields a parameterization of a classical-mechanical equation of motionthat describes the trajectory of the system, which is a spacetime individ-ual in the sense of the EMMO. In the same way, molmod:NH3 SITE COM ,which would represent the centre of mass of the molecule, cannot be asign in the sense of the EMMO: The centre of mass is a point (not avolume), and its trajectory over time is one-dimensional rather than four-dimensional. However, molmod:NH3 SITE COM is part of a sign, by belongingto molmod:NH3 RIGID UNIT . The procedure from Section 3.2 will now be followed to create the fragment ofan ontology alignment between marketplace-level ontologies and the EMMO.By reexpressing the source-ontology (OSMO, VISO, and VOV) scenario fromSection 2.3 in the target ontology (EMMO) as in Section 4.1, classes andrelations are mapped to each other (step 1) σ ◦ i τ ◦ i for 1 ≤ i ≤ , (1) emantic interoperability based on the EMMO 17 where the initial source terms are σ ◦ = osmo:einecs listed material , (2) σ ◦ = osmo:materials relation ,σ ◦ = σ ◦ = viso-am:rigid object ,σ ◦ = viso-am:mie site ,σ ◦ = viso-am:mass site ,σ ◦ = viso-am:structureless object ,σ ◦ = viso:has part ,σ ◦ = vov:involves ,σ ◦ = osmo:has aspect paradigmatic content , and the initial target terms are τ ◦ = emmo-material:material , (3) τ ◦ = emmo-models:material relation ,τ ◦ = emmo-semiotics:sign ,τ ◦ = emmo-graphical:symbolic ,τ ◦ = τ ◦ = τ ◦ = emmo-graphical:symbol ,τ ◦ = emmo-mereotopology:has proper part ,τ ◦ = ( emmo-mereotopology:has proper part ) − ◦ emmo-mereotopology:has proper part ,τ ◦ = ( emmo-models:has model ◦ emmo-mereotopology:has proper part ) − . Therein, p − denotes the inverse relation to p , and p ◦ q denotes the chainproduct of the relations p and q .By evaluating the respective candidate correspondences (step 2) on thebasis of the definitions from the marketplace-level ontologies and the EMMO, σ ◦ i ⊑ τ ◦ i is found to be valid for i ∈ { , , , , , } ; hence, σ ′ i = σ ◦ i and τ ′ i = τ ◦ i for these i . For i = 3, no non-trivial and valid relaxation can be found: Arigid object need not be a sign according to the EMMO, and whether it isone or not depends on its understanding by an interpreter. For the associatedsemiotic processes, a clear and simple rule cannot be formulated; accordingly,this candidate is discarded. The correspondences between relations can berelaxed by τ -generalization to τ ′ = emmo-mereotopology:has part , (4)with σ ′ = σ ◦ , and by σ -refinement to σ ′ = vov:involves ⊓ [ osmo:materials relation ] • (5) ⊓ • [ viso:model object ] ,σ ′ = osmo:has aspect paradigmatic content ⊓ [ osmo:materials relation ] • ⊓ • [ evmpo:material ] , with τ ′ = τ ◦ and τ ′ = τ ◦ , where [ c ] • is the relation that holds wheneverthe subject is an individual of class c (irrespective of the object), • [ c ] is therelation that holds whenever the object is an individual of class c (irrespectiveof the subject), and ⊓ is the intersection operator, applied to relations. Inthis way, σ ′ i ⊑ τ ′ i is valid for all the remaining cases ( i.e. , i = 3).Strengthening the correspondences (step 3) yields evmpo:material ≡ emmo-material:material (6)by σ -generalization and σ - τ -identification for i = 1, osmo:materials relation ≡ emmo-models:material relation (7)by σ - τ -identification for i = 2, viso:model object ⊑ emmo-graphical:symbolic (8)by σ -generalization for i = 4, viso-am:structureless object ⊑ emmo-graphical:symbol (9)for i ∈ { , } by σ -generalization and for i = 7 unchanged, viso:has part ⊑ emmo-mereotopology:has part , (10)for i = 8 unchanged, (cid:16) vov:involves (11) ⊓ [ osmo:materials relation ] • ⊓ • [ viso:model object ] (cid:17) ⊑ (cid:16)(cid:16) [ emmo-models:model ] • ⊓ emmo-mereotopology:has proper part (cid:17) − ◦ emmo-mereotopology:has proper part (cid:17) , for i = 9 by τ -refinement, and (cid:16) osmo:has aspect paradigmatic content (12) ⊓ [ osmo:materials relation ] • ⊓ • [ evmpo:material ] (cid:17) emantic interoperability based on the EMMO 19 ⊑ (cid:16) emmo-models:has model ◦ emmo-mereotopology:has proper part (cid:17) − , for i = 10 unchanged.Eqs. (6) to (10) can be expressed in OWL (step 4): evmpo:material owl:equivalentClass emmo-material:material.osmo:materials_relation owl:equivalentClass emmo-models:material_relation.viso:model_object rdfs:subClassOf emmo-graphical:symbolic.viso-am:structureless_object rdfs:subClassOf emmo-graphical:symbol.viso:has_part rdfs:subPropertyOf emmo-mereotopology:has_part. Therein, the first statement was already included in the EVMPO-EMMOintegration, cf.
Section 2.1 and Fig. 1, whereas the subsequent statementscontribute to aligning the marketplace-level ontologies with the EMMO.The present example also illustrates the limitations due to using OWLfor the purpose of formalizing the alignment. Eqs. (11) and (12) exceedthe expressive capacity of OWL; however, the meaning of these correspon-dences is intuitively straightforward, and similar cases occur often in prac-tice: Eq. (11), from i = 9, states that “if :X vov:involves :Y , where :X isa osmo:materials relation and :Y is a viso:model object , then thereis an emmo-models:model that relates to both of them by emmo-mereo - topology:has proper part ,” i.e. , they are both part of the same model.Eq. (12), from i = 10, expresses the rule that “if :X osmo:has aspect para - digmatic content :Y , where :X is a osmo:materials relation and :Y is a evmpo:material , then there is an individual :Z such that :Y em - mo-models:has model :Z and :Z emmo-mereotopology:has proper part:X .” Statements like these can easily be formulated in first-order logic, ex-tensions of OWL DL by additional operators (see above), or by graph trans-formation systems [24] which would apply to graph representations of thescenarios (see the Appendix). Data technology for CME/ICME services and platforms needs to aim atFAIR data management [7] and semantic interoperability with respect toconcrete software and data infrastructures, e.g. , materials modelling market-places, for which it is crucial to develop marketplace-level domain ontologiesby which the associated models, tools, infrastructures, and workflows, can bedocumented in detail. The EMMO, on the other hand, aims at capturing ma-terials modelling and characterization in general; beyond this, it is expectedto serve as an entirely domain-independent top-level ontology. The presentwork shows that the gap between the top level and the domain level can indeed be bridged, even where, as it is the case here, the involved ontologiesare at a relatively early stage of development.An ontology alignment, once established, becomes community knowledgeand remains available for future use; we suggest to include these align-ment statements in a TTL file that should be distributed together with theEVMPO and the marketplace-level ontologies [19]. As such knowledge is ac-cumulated over time, the EMMO becomes more accessible to its prospectivecommunity of users, not all of whom can be expected to familiarize them-selves with all intricacies of the philosophical conceptualization underlyingthe EMMO classes and relations. Domain ontologies, however, are compara-bly close to the language employed in a particular field and the intuition of do-main experts. Therefore, ontology matching, even fragment by fragment, suc-cessively contributes to creating the necessary link between domain-specificexpertise and the EMMO as an overarching interoperability standard.To illustrate this, it is improbable that among developers of molecu-lar simulation software, a majority will at any point be aware of the cir-cumstance that a Lennard-Jones interaction site can be documented as an emmo-graphical:symbol individual, or that a multi-site rigid-unit part ofa molecular model is best represented as an emmo-graphical:symbolic . In-stead, a CME/ICME domain expert might find it more easy to recognize that viso-am:lj site and viso-am:rigid object , respectively, are appropriateconcepts. Then, from VISO [20] and the present ontology alignment, viso-am:lj site ⊑ viso-am:structureless object (13) ⊑ emmo-graphical:symbol , viso-am:rigid object ⊑ viso:model object (14) ⊑ emmo-graphical:symbolic , the EMMO representation of these entities can be deduced correctly.To make this approach and the EMMO-based semantic interoperabilityarchitecture as a whole more viable, future work should address mechanismsby which alignment rules can be included if they are beyond the expressivecapacity of OWL DL; e.g. , graph rewriting rules could be applied to represen-tations of scenarios as graphs, building on the work in progress documentedin the Appendix. Furthermore, a community-approved procedure needs to beestablished for suggesting, evaluating, and disseminating ontology alignmentsthat connect the EMMO to domain knowledge. Acknowledgements
The authors thank N. Adamovic, W. L. Cavalcanti, ˚A. Ervik,A. Hashibon, and E. A. M¨uller for fruitful discussions. The co-authors Y.B., E.G.,G.G., G.M., and G.J.S. acknowledge funding from the European Union’s Horizon2020 research and innovation programme under grant agreement no. 723867 (EMMC-CSA), the co-authors Y.B., G.G., G.M., and G.J.S. under grant agreement no. 760173(MarketPlace), the co-authors S.C., G.G., M.T.H., G.M. under grant agreement no.emantic interoperability based on the EMMO 21760907 (VIMMP), and the co-author E.G. under grant agreement no. 814492 (Sim-DOME).
Appendix: Graph representation of EMMO scenarios
Using the EMMO Python API [12] in combination with the Lucidchart onlinetool [27], scenarios represented as a graph ( e.g. , see Fig. 2) can be convertedautomatically to a list of Python classes consistent with the Owlready2 format[25] which facilitates further processing by the EMMO API. The completeprocedure involves performing the following steps:1. Assign numerical labels to concepts and relations ( e.g. , see Tab. 1).2. Export the flowchart data from Lucidchart [27] into CSV format.3. Insert parameters in the parameters.py module file [12].4. Execute the
CSP ontology.py script [12] to generate the list of classes forOwlready2 [25].For further details, we refer to the EMMO Lucidchart documentation [33]and the upcoming release version of the EMMO.
Table 1
Codes (edge labels) employed for the EMMO relations in the present graphrepresentation.code EMMO relation name and description, cf.
Ghedini et al. [12]0 emmo-properties:has property a property is a “sign that stands for an object that the ’interpreter’perceived through a well defined observation process” [12], e.g. , anexperimental measurement process; has property relates the observedobject to the outcome of that process1 emmo-processual:has proper participant relates a process to an entity that participates in the process; irreflexiverelation (as opposed to has participant )2 emmo-semiotics:has sign relates an object to a sign that refers to it by participation in semiosis3 emmo-mereotopology:has proper part irreflexive parthood relation4 emmo-mereotopology:has spatial part “relation that isolates a proper part that extends itself in time within thelifetime of the whole, without covering the full spatial extension of the 4Dwhole ( i.e. is not a temporal part)” [12]2 Horsch, Chiacchiera, Bami, Schmitz, Mogni, Goldbeck, Ghedini Fig. 2
Graph representation of EMMO individuals and relations from a scenariodescribing a simulation workflow for crystal structure prediction; cf.
Tab. 1 for theedge labels. This diagram was created using Lucidchart [27].emantic interoperability based on the EMMO 23
References
1. Allemang, D., Hendler, J.: Semantic Web for the Working Ontologist, 2 edn.Morgan Kaufmann, Waltham, Massachusetts, USA (2011)2. Arnold, P., Rahm, E.: Enriching ontology mappings with semantic relations. DataKnowl. Eng. , 1–18 (2014)3. Arp, R., Smith, B., Spear, A.D.: Building Ontologies with Basic Formal Ontology.MIT Press, Cambridge, Massachusetts, USA (2015)4. Asuncion, C.H., van Sunderen, M.J.: Pragmatic interoperability: A systematicreview of published definitions. In: P. Bernus, G. Doumeingts, M. Fox (eds.)Enterprise Architecture, Integration and Interoperability, pp. 164–175. Springer,Heidelberg, Germany (2010)5. Baader, F., Horrocks, I., Lutz, C., Sattler, U.: An Introduction to DescriptionLogic. Cambridge University Press, Cambridge, UK (2017)6. Berto, F., Plebani, M.: Ontology and Metaontology. Bloomsbury, London, UK(2015)7. Bicarregui, J.: Building and sustaining data infrastructures: Putting policy intopractice. Tech. rep. (2016). DOI 10.6084/m9.figshare.4055538.v28. Borgo, S., Masolo, C.: Ontological foundations of DOLCE, pp. 279–295. Springer,Dordrecht, Netherlands (2010)9. de Buyl, P., Colberg, P.H., H¨ofling, F.: H5MD: A structured, efficient, andportable file format for molecular data. Comp. Phys. Comm. , 1546–1553(2014)10. Ceusters, W.: An information artifact ontology perspective on data collectionsand associated representational artifacts. Stud. Health Technol. Inform. (3), 97–101 (2013)18. Horsch, M.T., Chiacchiera, S., Seaton, M.A., Todorov, I.T., Schembera, B., Kon-chakova, N., Klein, P.: Reliable and interoperable computational molecular en-gineering: 1. Pragmatic interoperability and translation of industrial engineeringproblems into modelling and simulation solutions. In: M.T. Horsch, G. Boc-cardo (eds.) Molecular and Mesoscopic Modelling in Chemical Engineering DataScience (2020). In preparation4 Horsch, Chiacchiera, Bami, Schmitz, Mogni, Goldbeck, Ghedini19. Horsch, M.T., Chiacchiera, S., Seaton, M.A., Todorov, I.T., ˇSindelka, K., L´ısal,M., Andreon, B., Kaiser, E.B., Mogni, G., Goldbeck, G., Kunze, R., Summer,G., Fiseni, A., Br¨uning, H., Schiffels, P., Cavalcanti, W.L.: Ontologies for theVirtual Materials Marketplace. KI – K¨unstl. Intell. (2020). Submitted; preprint:arXiv:1912.01519 [cs.DB]20. Horsch, M.T., Niethammer, C., Boccardo, G., Carbone, P., Chiacchiera, S., Chiri-cotto, M., Elliott, J.D., Lobaskin, V., P. Neumann, P.S., Seaton, M.A., Todorov,I.T., Vrabec, J., Cavalcanti, W.L.: Semantic interoperability and characterizationof data provenance in computational molecular engineering. Fluid Phase Equilib.(2020). DOI 10.1021/acs.jced.9b0073921. Hu, B., Hu, B.: Tower of Babel: Interoperability of ontologies for pervasive com-puting. In: First International Symposium on Pervasive Computing and Appli-cations, pp. 690–695. IEEE, Piscataway, New Jersey, USA (2006)22. Iliadis, A.: The Tower of Babel problem: Making data make sense with BasicFormal Ontology. Online Informat. Rev. (6), 1021–1045 (2019)23. Jim´emez Ruiz, E.: Ontology Alignment Evaluation Initiative (OAEI).http://oaei.ontologymatching.org/ (2019). Date of access: 31st December 201924. K¨onig, B., Nolte, D., Padberg, J., Rensink, A.: A tutorial on graph transforma-tion. In: R. Heckel, G. Taentzer (eds.) Graph Transformation, Specifications, andNets, pp. 83–104. Springer, Cham, Switzerland (2018)25. Lamy, J.B.: Owlready: Ontology-oriented programming in Python with auto-matic classification and high level constructs for biomedical ontologies. Artif.Intell. Med. , 11–28 (2017)26. Lewis, D.: New work for a theory of universals. Australas. J. Phil. (4), 393–412 (2012)29. Mascardi, V., Cord`ı, V., Rosso, P.: A comparison of upper ontologies. In: M. Bal-doni, A. Boccalatte, F. De Paoli, M. Martelli, V. Mascardi (eds.) WOA 2007:Dagli Oggetti agli Agenti. Agenti e Industria: Applicazioni tecnologiche degliagenti software, pp. 55–64. Seneca Edizioni, Torino, Italy (2007)30. Mascardi, V., Locoro, A., Rosso, P.: Automatic ontology matching via upperontologies: A systematic evaluation. IEEE Transact. Knowl. Data Eng. (5),609–623 (2010)31. Merkys, A., Mounet, N., Cepellotti, A., Marzari, N., Graˇzulis, S., Pizzi, G.: Aposteriori metadata from automated provenance tracking: Integration of AiiDAand TCOD. J. Cheminform. , 56 (2017)32. Miller, G.A.: WordNet: A lexical database for English. Comm. ACM (11),39–41 (1995)33. Mogni, G.: Tutorial on converting Lucidchart ontology flowcharts to Owlready2Python format (2019)34. Noy, N.F.: Semantic integration: A survey of ontology-based approaches. SIG-MOD Rec. (4), 65–70 (2004)35. Peirce, C.S.: Peirce on Signs: Writings on Semiotic. University of North CarolinaPress, Chapel Hill, USA (1991)36. Pickel, B., Mantegani, N.: A Quinean critique of ostrich nominalism. Philos.’Impr. (6), 1–21 (2012)37. Schembera, B., Dur´an, J.M.: Dark data as the new challenge for big data scienceand the introduction of the scientific data officer. Philos. Technol. (2020). DOI10.1007/s13347-019-00346-x38. Schmitz, G.J.: Microstructure modeling in integrated computational materialsengineering (ICME) settings: Can HDF5 provide the basis for an emerging stan-dard for describing microstructures? JOM (1), 77–83 (2016)emantic interoperability based on the EMMO 2539. Schmitz, G.J., B¨ottger, B., Apel, M., Eiken, J., Laschet, G., Altenfeld, R., Berger,R., Boussinot, G., Viardin, A.: Towards a metadata scheme for the description ofmaterials: The description of microstructures. Sci. Technol. Adv. Mater. (1),410–430 (2016)40. Schoop, M., de Moor, A., Dietz, J.: The pragmatic web: A manifesto. Comm.ACM (5), 75–76 (2006)41. Shvaiko, P., Euzenat, J.: Ontology matching: State of the art and future chal-lenges. IEEE Transact. Knowl. Data Eng. , 259–286 (1996)45. White, T.O.H., Murray-Rust, P., Couch, P.A., Tyer, R.P., Bruin, R.P., Todorov,I.T., Wilson, D.J., Dove, M.T., Austen, K.F., Parker, S.C.: Application and usesof CML within the eMinerals project. In: S.J. Cox (ed.) Proceedings of theUK e-Science All Hands Meeting 2006, pp. 606–613. National e-Science Centre,Edinburgh, UK (2006)46. Wi´sniewski, R., Wi´sniewska, M., Jarnut, M.: C-exact hypergraphs in concurrencyand sequentiality analyses of cyber-physical systems specified by safe Petri nets.IEEE Access , 13510 – 13522 (2019)47. Xu, X., Range, J., Gygli, G., Pleiss, J.: Analysis of thermophysical properties ofdeep eutectic solvents by data integration. J. Chem. Eng. Data (2019). DOI10.1021/acs.jced.9b0055548. Zhang, X., Li, K., Zhao, C., Pan, D.: A survey on units ontologies: Architecture,comparison and reuse. Program Electron. Lib. (2), 193–213 (2017)49. Zhao, S., Qian, Q.: Ontology based heterogeneous materials database integrationand semantic query. AIP Adv.7