CHARACTERIZING SEMANTIC SERVICE PARAMETERS
WITH ROLE CONCEPTS TO INFER DOMAIN-SPECIFIC
KNOWLEDGE AT RUNTIME
Alban Gaignard, Johan Montagnat
CNRS / UNS, I3S lab, MODALIS team, Sophia-Antipolis, France
Bacem Wali, Bernard Gibaud
INSERM / INRIA / CNRS / Univ. Rennes 1, IRISA Unit VISAGES U746, Rennes, France
Keywords:
Semantic web services, Role modeling, Reusable inference rules, Scientific workflows.
Abstract:
E-Science platforms leverage Service Oriented Architecture (SOA) principles to deliver large catalogs of data
processing services and experiments description workflows. In spite of their growing success, the usability of
these platforms is hampered by their catalogs size and the domain-specific knowledge needed to manipulate
the services provided. Relying on domain ontologies and semantic services to enhance the understanding and
usability of e-Science platforms, our contribution is twofold. First, we propose to delineate role concepts
from natural concepts at domain ontology design time which leads to a neuroimaging role taxonomy, making
explicit how neuroimaging datasets are related to the data analysis services. Then we propose to exploit,
at workflow runtime, provenance information extended with these domain roles, to infer new meaningful
semantic annotations. Platform semantic repositories are thus transparently populated, with newly inferred
annotations, through the execution of e-Science workflows. A concrete example in the area of neurosciences
illustrates the use of role concepts to create reusable inference rules.
1 INTRODUCTION
Semantically representing information has become a
de facto technique to enrich e-Science experimental
platforms with domain-specific knowledge. This ap-
proach aims at facilitating platforms usage, sharing of
experimental data and results, and experiments them-
selves, to finally foster collaborations among large
user groups. Conceptualizing domain knowledge, on-
tologies became a cornerstone for the underlying In-
formation Systems, as they are built upon controlled
vocabularies, logical constraints and inference rules.
Generally relying on Service Oriented Architec-
tures (SOA), e-Science experimental platforms pro-
vide tools dedicated to the publication, the identifi-
cation, and the invocation of data processing services.
However the technical description of services (e.g. us-
ing WSDL) does not provide any understanding on
the nature of the information processed nor on the op-
erations applied. Exploiting catalogues of data proce-
ssing services, e.g. to design flows of services (work-
flows) requires a clear understanding of how data is
processed and the nature of the data transformation
implemented by the services. Today, users are ex-
pected to have acquired this knowledge, which limits
the platforms usability to a restricted number of ex-
perts.
In this context, and relying on ontologies, seman-
tic (web) services tend to explicit the understanding of
(i) the nature of processed data and (ii) the nature of
the information processing applied to benefit, both at
experiment design-time and runtime, from the knowl-
edge on the services manipulated. Different levels of
semantic information can be distinguished:
1. Generic information, related to the technical de-
scription of services (e.g. semantic service de-
scriptions based on OWL-S) or related to the ser-
vice invocation which can later be used to produce
provenance traces (e.g. following the Open Prove-
nance Model).
59
Gaignard A., Montagnat J., Wali B. and Gibaud B..
CHARACTERIZING SEMANTIC SERVICE PARAMETERS WITH ROLE CONCEPTS TO INFER DOMAIN-SPECIFIC KNOWLEDGE AT RUNTIME.
DOI: 10.5220/0003648200590070
In Proceedings of the International Conference on Knowledge Engineering and Ontology Development (KEOD-2011), pages 59-70
ISBN: 978-989-8425-80-5
Copyright
c
2011 SCITEPRESS (Science and Technology Publications, Lda.)
2. Domain-specific information related to the nature
of the information processing realized by a service
invocation and the nature of the data manipulated
(e.g. taxonomies describing the nature of Dataset
and Dataset-processing). This knowledge can be
used to validate service invocations, by ensuring
that the expected types of data are used when in-
voking a service.
3. Domain-specific information related to the Role
played by the data involved in the service execu-
tion, from the service point of view. This knowl-
edge is needed both to ensure coherency of service
invocations, and to reason on the service invoca-
tion effect on the data processed.
Leveraging existing ontologies to describe generic
information as well as domain-specific nature of data
and processing tools, this paper focuses on the third
level of semantic information. The proposed ap-
proach tackles 3 aspects of semantic services manip-
ulated:
It clarifies the bindings between service descrip-
tions and domain concepts through a taxonomy of
domain-specific Roles.
It enables the coherency of service workflows de-
sign.
It makes it possible to infer new knowledge along
platform exploitation. This last point is achieved
by describing reusable domain-specific knowl-
edge inference rules associated to specific na-
tures of processing. The application of such rules
on a semantic database containing traces of ser-
vices invocation enriches the experimental plat-
forms with new valuable expert information.
We rely on the NeuroLOG platform (Montagnat
et al., 2008) to implement the concepts and support
experiments reported in this work. NeuroLOG is a
distributed environment designed to support the setup
of multi-centric studies in neurosciences. The On-
toNeuroLOG ontology (Temal et al., 2008) was de-
signed in the context of the platform development to
enhance the sharing of neuroimaging data and associ-
ated data analysis services.
The remainder of this paper is organized as fol-
lows. Section 2 presents some background on role
modeling and how it could be related to existing ini-
tiatives in the Semantic Web Services area. Section 3
motivates our approach through a small neuroimaging
workflow example. Ontologies supporting our work
are briefly presented in section 4. The benefits of re-
lying on Role concepts when designing a domain on-
tology are exposed in section 5, followed with sec-
tion 6 briefly illustrating how we complemented our
workflow environment with semantic web technolo-
gies. We finally discuss and conclude our approach in
section 7.
2 BACKGROUND INFORMATION
2.1 Role Modeling
In conceptual modeling, it is now agreed to sepa-
rate several categories of concepts, for instance those
characterizing the nature of an entity from those char-
acterizing their relations to each others. Henriksson
et al. propose a methodology based on the design of
role-based ontologies, extending standard ontologies,
to enhance ontology modularization and reusability.
They promote a clear delineation between Natural
Types and Role Types (Henriksson et al., 2008) : “In
role modeling, concepts that can stand on their own
are called natural types, while dependent concepts
are called role types”. Sowa (Sowa, 1984) first in-
troduced Natural Types to describe what is essential
to the identity of an individual, and Role Types to de-
scribe temporal or accidental relations to other indi-
viduals. The methodology proposed by Henriksson et
al. consists in (i) identifying the natural types of the
domain, (ii) identifying accidental or temporary rela-
tionships between individuals and ensuring that role
models are self-contained (for reusability) and finally
(iii) defining bridge axioms to bind role types to natu-
ral types (or to link individuals through properties de-
fined in the role model). This approach is particularly
interesting in our context since in Life Science ontolo-
gies, the design effort generally focuses on the first
step. Moreover, e-Science experimental platforms are
generally data-driven and well supported by ontolo-
gies describing the nature of data. But few efforts
concentrate in making explicit the knowledge relat-
ing data to their analysis services more deeply than
just using information on data nature.
2.2 Web Service Ontologies
Semantically enhanced e-Science experimental plat-
forms usually rely on standard generic service ontolo-
gies to describe data analysis services. The following
paragraphs briefly describe major service ontologies
and how they consider relations between data and ser-
vices.
The OWL-S Profile ontology (Martin et al., 2007),
one of the three ontologies forming the OWL-S pro-
posal, aims at describing what the annotated service
does. The service Profile presents a high-level inter-
face of the service through the properties hasParam-
KEOD 2011 - International Conference on Knowledge Engineering and Ontology Development
60
eter, hasInput, hasOuput. These properties link the
OWL-S Profile ontology to the OWL-S Process ontol-
ogy (aimed at describing the service internal behav-
ior) which defines the service parameters (Parameter
class) and their subclasses (Input and Output classes).
The type of parameters is given through the param-
eterType property of the Process ontology and spec-
ifies the classes/datatypes the value of the parameter
belongs to. According to the OWL-S specification,
nothing is said regarding how these parameter val-
ues are related to the service process and as a con-
sequence, these types should be considered as natu-
ral types as they are defined by Sowa (Sowa, 1984).
To specify the relationship of parameter values to the
process, it should be beneficial to rely, through the
parameterType property, on a role ontology designed
according to the methodology proposed by Henriks-
son et al..
FLOWS (Gruninger et al., 2008; Battle et al.,
2005) specifies a first-order logic ontology for Web
Services. It aims at enabling reasoning on the seman-
tics of services and their interactions. FLOWS has
largely been influenced by OWL-S but in addition,
it addresses interactions with business process indus-
try standards such as BPEL. FLOWS differs from
OWL-S by properly handling messages as core con-
cepts. Messages are defined in FLOWS by a mes-
sage type, characterizing the type of the content, and
a payload, the content itself. FLOWS defines also
three relations to relate atomic process invocations
to messages they consume as input or they produce
as output: produces, reads, and destroy message.
The relations are very generic and do not charac-
terize more precisely the consumption/production of
messages through domain-specific entities. However,
FLOWS proposes the described by relation to asso-
ciate a fluent to a message. Fluents are used to model
“changing” parts of the world. The described by rela-
tion aims at providing information on how the content
of the message impact the service invocation while
consuming/producing it. Intuitively, since role types
are defined by Sowa as accidental (or evolving during
time) relationships between entities, FLOWS’s fluents
could be a way to model how data are interpreted by
analysis services through Roles.
WSMO (Roman et al., 2006) is based on Orches-
tration to describe the internal behavior of services,
and on Choreography to describe their external be-
haviors. The Choreography of a service is described
through the importation of a domain ontology, which
defines the choreography state signature. This sig-
nature specifies, among other things, the service in-
puts and outputs as instances of the imported ontol-
ogy. WSMO is a rich service modeling and enacting
framework but it does not cover precisely the charac-
terization of how processed or produced data are re-
lated to services in terms of roles. Relying on external
ontologies, WSMO service interface remains compat-
ible with any domain ontology designed using a clear
separation between natural types and role types.
SAWSDL (Kopeck
´
y et al., 2007) is the W3C rec-
ommendation to semantically annotate WSDL and
XML Schema documents specifying standard Web
Services. These documents are bound to semantic en-
tities through the modelReference XML attribute. The
value assigned to a modelReference comprises a set
of zero or more URIs identifying concepts in an on-
tology. Again, this specification does not bring any-
thing new to separate the natural type of the annotated
WSDL message from how it is related to the Web
Service (its role type). However, depending on the
availability of an ontology of roles, modelReference
attributes could be used to bind role types to service
parameters.
Originating from the WSMO initiative, WSMO-
Lite (Vitvar et al., 2008) is built upon SAWSDL and
is a lightweight bottom-up approach, to semantically
describe Web Services and to enable reasoning on (i)
their associated semantic annotations, and (ii) their
interactions. Since WSMO-Lite uses SAWSDL to
bridge domain-specific ontologies with the service
description, roles types should be considered as an ex-
ternal feature, coming from the design of the domain
ontology.
2.3 Semantic Workflow Environments
Being based on either standard service ontologies, or
home-made approaches, the following paragraphs de-
scribe initiatives aiming at enhancing service discov-
ery, in the context of workflow environments.
The METEOR-S (Sheth et al., 2008) research
project is a major initiative in the Semantic Web Ser-
vices area. The approach is based on a peer-to-peer
middleware to address service discovery and publi-
cation. SAWSDL is used for both services annota-
tion (through modelReferences) and data mediation
(through schema lifting/lowering).
Built upon the
my
Grid ontology (Wolstencroft
et al., 2007), a bioinformatics service and domain
ontology, FETA (Lord et al., 2005) is a service dis-
covery framework characterized by a light-weight se-
mantic support and a semi-automatic approach. Three
main actors are distinguished in this framework: both
knowledge engineers and service annotators provide
semantic enhanced web services consumed by scien-
tists. Also built upon the
my
Grid ontology, the BioCat-
alogue (Bhagat et al., 2010) initiative is a community-
CHARACTERIZING SEMANTIC SERVICE PARAMETERS WITH ROLE CONCEPTS TO INFER
DOMAIN-SPECIFIC KNOWLEDGE AT RUNTIME
61
driven, and curated service registry aiming at guiding
users into a jungle of web services through the regis-
tration and annotation of web services and the brows-
ing of resulting annotated web services. Several kinds
of annotations are available going from free text, to
tags or ontology terms. BioCatalogue allows, among
other kind of annotations to operationally (e.g. infras-
tructure, runtime constraints) or functionally describe
a service. Functional annotation covers information
related to what the service does, but also its function
and the format of input or output data. The function
annotation of data with regard to a given web service
seems to be close to Role types previously introduced
but few information is available to precisely describe
this kind of annotation.
The BioMOBY project aims at providing interop-
erability for biological data centers and analysis cen-
ters. SAWSDL has been used in this context and this
real-world application is one of the few existing ini-
tiatives (Gordon and Sensen, 2008). The focus is on
interoperability and therefore on schema mapping an-
notations of SAWSDL, implemented through XSLT
stylesheets. The entry-point is a SAWSDL Proxy
servlet, in front of a web service provider, a semantic
registry, and a schema mapping server. As a continua-
tion of this initiative, the SADI project (Withers et al.,
2010) proposes guidelines and best-practices to en-
hance semantic service discovery at workflow design
time. Semantic services are indexed in the catalog
through the new set of RDF properties describing the
resulting new semantic features associated to input
data. The service discovery is based on searches over
data properties consumed as input and over the pro-
duced new properties. This approach also aims at re-
ducing ambiguity of search queries through more pre-
cise properties, describing the relationships between
input and output data. We propose to address such re-
lationships at domain ontology design time, through
a taxonomy of Roles, clearly identifying the role of
data with regard to their analysis services.
3 MOTIVATING USE CASE
3.1 Image Registration Workflow
The workflow illustrated in Figure 1 represents a typi-
cal image registration process commonly encountered
in neurosciences workflows. It consists in superim-
posing two medical images acquired independently
into the same coordinate system. The sample regis-
tration process is decomposed into two steps. First,
the registration itself consists in calculating, from the
input brain MRI and a brain atlas, a geometrical trans-
Registration Re-sampling
100
b
tz
a
yx
MRI
MRI
MRI
Matrix
Figure 1: A typical neuroimaging workflow mixing several
nature of data and processing.
formation expressed by a transformation matrix. Sec-
ond, the resampling step effectively aligns the input
brain MRI by applying the transformation expressed
through the registration matrix.
In spite of its apparent simplicity, this workflow is
interesting for the following reasons. First, this work-
flow mixes two services of different nature, whose
meaning has been agreed upon within the image pro-
cessing community. In other words, the knowledge
about what kind of underlying treatment is clear for
the community and is generally not explicit at the
tooling level. Second, this workflow consumes and
produces data of several natures (medical images,
transformation matrix) expressed through raw files at
the tooling level. Again, these files have a precise
meaning from the user community point of view, with
regard to their content. Finally, the first step of the
workflow takes two files as input, sharing the same
nature, both are brain MRIs, but they play different
roles from the processing tool perspective. The first
one is used as the reference image for the registration
process (atlas) whereas the second one is used as the
floating (moving) image. This knowledge is hidden
at the tooling level, and even for domain experts, the
variability of tools makes their configuration not triv-
ial.
3.2 Enriching the Semantic Repository
with Valuable Annotations
Relying on a semantic data repository together with a
reasoning engine, we consider in this paper a method-
ology for producing and deducing new meaningful
facts from the user community perspective. For ex-
ample, considering the result of the registration work-
flow presented in Figure 1, it should be interesting to
retrieve, the atlas used in the registration. More gen-
erally, our approach tends towards the propagation of
the effect of a service (or a sub-part of the workflow)
to the produced data. For instance, we would like
to automate the generation of a fact saying that “this
dataset can be superimposed with this dataset”, be-
cause some processing tools might require that their
inputs have beforehand been registered.
KEOD 2011 - International Conference on Knowledge Engineering and Ontology Development
62
Registration Re-sampling
100
b
tz
a
yx
wasGeneratedBy
used
used
used
used
wasGeneratedBy
can-be-superimposed-with
Figure 2: Linking data and processes through generic and
domain-specific relations.
Figure 2 illustrates the semantic relations estab-
lished between entities involved in the sample work-
flow, i.e. data and services. Black arrows are rela-
tions that can be created on-the-fly during each ser-
vice invocation. It states data production and con-
sumption knowledge. Beyond linking together data
and processes (i.e. capturing provenance informa-
tion), we want to rely on both an ontology and a
reasoning engine to infer relevant domain informa-
tion using rules expressing domain knowledge. For
instance the dashed arrow represents such informa-
tion derived using a domain-specific rule embedding
domain knowledge about the overall registration pro-
cess.
As it will be shown in Section 5, this kind of
knowledge inference is possible if the services seman-
tic description is rich enough to properly define the
Roles of processed data in the context of services in-
vocation. In addition, such high-level semantic de-
scription can be used to validate the coherency of
flows of services. Before entering into the details, the
next Section describes the ontologies on which this
work is grounded.
4 SUPPORTING ONTOLOGIES
4.1 Domain Ontology
The OntoNeuroLOG ontology was developed to
provide common semantics for information sharing
throughout the NeuroLOG system. Indeed, the ulti-
mate goal of NeuroLOG was to allow the success-
ful sharing of neuroimaging resources provided by
collaborating actors in the field of neuroimaging re-
search, the term resources covering both neuroimag-
ing data (such as images) as well as image processing
programs, shared as services. This ontology is used as
a reference to query and retrieve heterogeneous data,
thanks to the mediation system, as well as to annotate
consistently the shared services, i.e. denote what sort
of processing such services actually achieve and what
data they accept as input and produce as result.
OntoNeuroLOG was designed as a multi-layer ap-
plication ontology, relying on a number of core on-
tologies modeling entities that are common to several
domains. The whole ontology relies on DOLCE (De-
scriptive Ontology for Language and Cognitive Engi-
neering), a foundational ontology that provides both
the basic entities (at the top of the entities’ taxonomy)
and a common philosophical framework underlying
the whole conceptualization. The ontology was de-
signed according to the OntoSpec methodology (Kas-
sel, 2005), which focuses on the writing of semi-
formal documents capturing rich semantics. This is
followed by an implementation of a subset of the on-
tology in OWL, the web ontology language. The
definition of this subset and the choice of the rele-
vant OWL dialect take into account the specific needs
of the application. Two subsets of OntoNeuroLOG
were used in the context of this work, the ontology of
Dataset and the ontology of Dataset processing, in-
troduced hereafter.
4.1.1 Dataset Sub-ontology
Datasets are Propositions (i.e. Non physical en-
durants) that represent the content of data files used
in neuroimaging. The taxonomy of Datasets is orga-
nized according to several semantic axes. The first
denotes what facet of the subject is explored, e.g.
Anatomical datasets explore the subject’s anatomy
whereas Metabolic datasets explore brain metabolic
processes. The second axis classifies Datasets ac-
cording to some imaging modality, such as Com-
puted Tomography (CT), Magnetic resonance (MR),
Positron emission tomography (PET). This axis in-
cludes the numerous sub-modalities met, e.g., in MR
imaging such as T1-weighted MR dataset, Diffusion-
weighted MR dataset, etc. The third axis focuses
on Datasets that result from some kind of post-
processing, such as Reconstructed datasets, Registra-
tion datasets, Segmentation datasets, etc.
Datasets may bear properties of Representational
objects (since Propositions are Representational ob-
jects), such as ’refers to’, which denotes the ability
to refer to any kind of Particular. This property can
be used to refer, e.g. to the Subject (i.e. the patient)
concerned by a particular Dataset. For instance, a
property called ’can be superimposed with’ was in-
troduced to relate two Datasets that can be superim-
posed with each other, such as a Segmentation dataset
(i.e. an object mask obtained through a segmentation
procedure) and the original dataset from which it was
obtained.
CHARACTERIZING SEMANTIC SERVICE PARAMETERS WITH ROLE CONCEPTS TO INFER
DOMAIN-SPECIFIC KNOWLEDGE AT RUNTIME
63
4.1.2 Dataset-processing Sub-ontology
Dataset processings are Conceptual actions (i.e. Per-
durants) that affect Datasets. The taxonomy of
Dataset processings covers the major classes of image
processing met in neuroimaging, such as: restoration,
segmentation, filtering, registration, re-sampling, etc.
Axioms attached to each Dataset processing class
usually denote which classes of Datasets are being
processed or result of the corresponding processing.
For example, a Reconstruction ’has for data’ some
Non-reconstructed dataset and ’has for result’ some
Reconstructed dataset; a Segmentation ’has for re-
sult’ some Segmentation dataset.
4.2 Ontology of Web Services
In addition, an ontology was defined to describe Web
Services grounded to the DOLCE foundational con-
cepts. It introduces the notions that are classically in-
volved in WS specifications such as the notions of in-
terface (ws-interface), operation (ws-operation), ser-
vice inputs and outputs (input/output-variable). Be-
sides, the model introduces a ’refers to’property to es-
tablish relationships with the classes of data process-
ing that a particular ws-operation implements (such
as rigid-registration or segmentation), as well as with
the classes (natural types) of entity that the input and
output variable actually represent.
4.3 OPM Ontology
The Open Provenance Model (Moreau et al., 2011)
initiative (OPM) aims at homogenizing the expres-
sion of provenance information on the wealth of data
produced by e-Science applications. Among other
things, OPM enables the exchange of provenance in-
formation between several workflow environments. It
eases the development of tools to process such prove-
nance information, and finally facilitates the repro-
ducibility of e-Science experiments.
OPM is materialized through a natural language
specification and three formal specifications: an XML
schema (OPMX), an OWL ontology (OPMO) and a
controlled vocabulary, with simpler OWL constructs
(OPMV). OPM defines directed graphs representing
causal dependencies between “things”. A Causal de-
pendency is defined as a directed relationship between
an effect (the source of the edge) and a cause (the des-
tination of the edge). The nodes of the provenance
graph might be either an Artifact (immutable, state-
less element), or a Process (actions performed on an
Artifact and producing new ones), or an Agent (en-
tity controlling or affecting the execution of a Pro-
cess). The edges of the graph represent (i) dependen-
cies between two artifacts (wasDerivedFrom) to track
the genealogy of artifacts, (ii) dependencies between
two processes (wasTriggeredBy) to track the sequence
of processes, and (iii) dependencies between artifacts
and processes (used/wasGeneratedBy) to track the
consumption and the production of artifacts through
processes. Additionally OPM allows to track the links
between processes and their enactor agents through
wasControlledBy dependencies.
However these kinds of dependencies are very
generic and are proposed as a basis to track input ar-
tifacts and output artifacts produced through process
invocations. To distinguish several causal dependen-
cies of the same kind, OPM allows to annotate used or
wasGeneratedBy dependencies with syntactic roles.
A Role is defined in OPM as a particular function of
an artifact (or an agent) in a process. The OPM model
does not formally define roles but allows to “tag”
dependencies between artifacts (or agents) and pro-
cesses with meaningful labels. In OPM the execution
of the sample registration process illustrated in Fig-
ure 1 could be translated with these two statements
Registration
Process
: used( f loating) : Image
Arti f act
and Registration
Process
: used(re f erence) :
Atlas
Arti f act
”. The syntactic roles f loating
and re f erence aim at distinguishing how artifacts
are related to processes, but their meaning remains
highly dependent on their usage within a given
process, and thus, remain highly domain-specific.
In the following section we propose to go deeper
with the notion of roles, through the proposition of
a taxonomy of Roles in neuroimaging that aims at
(i) enhancing the semantic annotation of services,
and (ii) exploiting OPM provenance information
to deduce meaningful statements in the context of
neuroimaging workflows.
5 ROLE CONCEPTS
To benefit from expert knowledge conceptualized
through a domain ontology (such as the OntoNeu-
roLOG ontology), services involved in e-Science
workflows are manually associated to concepts of
the ontology. Semantically annotating a service con-
sists in using an ontology to bind technical concepts,
i.e. elements syntactically describing services, to
domain-specific concepts. Most of semantic web-
services initiatives, namely OWL-S, WSMO, SWSO,
or SAWSDL, distinguish the annotation of the func-
tionality of the service from the annotation of the ser-
vice parameters which consume or produce data. For
instance, let us consider a medical image processing
KEOD 2011 - International Conference on Knowledge Engineering and Ontology Development
64
Figure 3: A Role taxonomy characterizing how neuroimaging data can be related to neuroimaging processing tools.
tool performing a de-noising operation. From a tech-
nical or syntactical point of view, the service might
be implemented by an executable binary taking as in-
put a raw file materializing a noised medical image
and producing as output another raw file materializ-
ing the resulting de-noised image. From a seman-
tic point of view, this de-noising service might im-
plement a particular kind of algorithm characterizing
how the image is processed. This “how” should be
described through the annotation of the functional-
ity of the service, i.e. a particular class of restoration
processing. The service might additionally require a
specific medical image format, and a specific modal-
ity of acquisition, for instance ultrasound. Moreover,
the resulting de-noised image should preserve the in-
put modality; in other words, even de-noised, the im-
age still remains an ultrasound image. The service in-
put/output parameters are usually annotated with con-
cepts describing the nature of consumed or produced
data. We will see in the following section that such se-
mantic annotation of the nature of consumed or pro-
duced data is often not sufficient to be precisely ex-
ploited to produce new domain-specific annotations.
5.1 Differentiating Natural and Role
Concepts
Service annotation should also make explicit how
consumed or produced data items are related to the
processes. For instance, if we consider the registra-
tion service involved in the workflow shown in Fig-
ure 1, both input parameters should share the same
intrinsic nature. Indeed, in this example, the refer-
ence image parameter and the floating image param-
eter have been acquired both through the same Mag-
netic Resonance modality (MR) and should be mate-
rialized with the same file format. In this geometrical
realignment procedure, the two input parameters are
not distinguished by their intrinsic nature but rather by
their relationship to the registration process, namely
floating and reference images. It is important to note
that these two concepts only make sense in the con-
text of a particular kind of image processing, registra-
tion. Without the knowledge of “which data is acting
as the reference image” or “which data is acting as the
floating image”, it is difficult to deduce any meaning-
ful information from the execution of the registration
workflow, such as “this resulting image can be super-
imposed with this reference image”, or more gener-
ally to retrieve images that have been registered with
the same reference and thus, that can be superimposed
together.
To tackle this issue we propose to distinguish Nat-
ural concepts and Role concepts when annotating se-
mantic service parameters by relying on a domain-
specific role taxonomy.
Figure 3 illustrates the taxonomy of roles dedi-
cated to the characterization of the relationships be-
tween neuroimaging data and their dedicated process-
ing. Role concepts are organized following the main
classes of neuroimaging processing like in the On-
toNeuroLOG dataset processing ontology.
Relying on this taxonomy of roles, we are now
able to precisely annotate the input and output pa-
rameters of our image registration service with both
Natural concepts and Role concepts. Both input im-
CHARACTERIZING SEMANTIC SERVICE PARAMETERS WITH ROLE CONCEPTS TO INFER
DOMAIN-SPECIFIC KNOWLEDGE AT RUNTIME
65
ages are characterized by a same Natural concept, T1
weighted magnetic resonance image (T1-MR). T1-
MR can be considered as a Natural concept because it
stands on its own and does not characterize how input
data are related with any other entities. On the other
hand, service input parameters can be annotated with
two distinct Role concepts to characterize how input
data are related to the registration process. The ser-
vice input parameter interpreting data as floating (the
moving data, that will finally be realigned) is anno-
tated with role As-floating-image, and the second ser-
vice input parameter interpreting data as the geomet-
rical reference is annotated with role As-reference-
image. Figure 4 illustrates the annotation with Role
concepts for the two services involved in the full reg-
istration use-case workflow.
Registration Re-sampling
100
b
tz
a
yx
As-resampled
As-unprocessed
As-floating
As-transformation
As-reference
As-transformation
Figure 4: Roles involved in the registration workflow.
Disambiguating the semantic annotation of ser-
vices, we present in the following section how Role
concepts are the basis to instrument domain ontolo-
gies with reusable inference rules, producing new
meaningful statements.
5.2 Integration of OPM and
OntoNeuroLOG
The Roles taxonomy also acts as a bridge ontology,
articulating the two technical ontologies dedicated to
the description of services (Web Services) and to the
provenance information associated to their invocation
(OPM). Indeed, Role concepts are associated to the
service I/Os (input/output-variable) through the same
property (refers-to) as used to describe the nature
of consumed/produced data (OntoNeuroLOG Dataset
ontology). Moreover, Role concepts are directly ex-
tending the OPM Role class, so that when recording
provenance at workflow runtime, the workflow enac-
tor is able to link Artifacts to Processes through the
newly refined Roles.
5.3 Reusable and Service Independent
Inference Rules
The use of rule engines (inference engines) is a well
adopted data-driven and declarative approach to de-
duce new conclusions and thus produce new facts
from a set of statements. In an OPM-instrumented
execution engine, the invocation of services generates
provenance statements such as the ones illustrated in
Figure 5 for the registration workflow. The graphical
syntax introduced by (Moreau et al., 2011) is reused:
Artifacts are represented by ellipses and Processes are
represented by rectangles, used and wasGeneratedBy
causal dependencies, parametrized with roles, are rep-
resented by plain arrows.
Registration
Image
Atlas
used
(As_floating_image)
used
(As_reference_image)
Matrix
wasGeneratedBy
(As_affine_transformation)
Resampling
Image
Matrix
used
(As_unprocessed)
used
(As_affine_transformation)
Result
wasGeneratedBy
(As_resampled_image)
Figure 5: OPM statements recorded through the invocation
of the registration workflow.
To automate the production of a statement linking
the resulting data to the source data through a domain-
specific property, an inference rule is written using the
role-parameterized provenance causal dependencies.
For instance, Figure 6 illustrates the inference rule
deducing the can be superimposed with property in
the case of the registration workflow. The left part of
the implication, the antecedent corresponds to the If
clause of the inference rule and consists in identify-
ing a conjunction of statements necessary to produce
the statements expressed in the consequent, the right
part of the implication (the Then clause of the rule).
The first two lines assert that processes must refer,
for the first one, to a Registration treatment, and for
the second one, to a Resampling treatment. In other
words, the services invoked by the processes should
have been annotated with the corresponding Natu-
ral concepts of the OntoNeuroLOG domain ontology.
The two following lines of the If clause identify arti-
facts and processes through their Role concepts: the
resulting image is identified through As-resampled-
image, the registration matrix is identified through
As-affine-transformation, and the reference image is
identified through As-reference-image. Finally, when
the reference image and the resulting resampled im-
age are identified, the rule engine is able to produce a
new statement saying that both images can be super-
imposed (can
be superimposed with property of the
OntoNeuroLOG ontology).
Using Role concepts, domain ontologies can be
instrumented with inference rules which remain ser-
vice independent. Such inference rules can be reused
KEOD 2011 - International Conference on Knowledge Engineering and Ontology Development
66
RegistrationAtlas
used
(As_reference_image)
Matrix
wasGeneratedBy
(As_affine_transformation)
ResamplingMatrix
used
(As_affine_transformation)
Result
wasGeneratedBy
(As_resampled_image)
Resampling Resampling-Class
Registration Registration-Class
Result
can_be_superimposed_with
Atlas
Figure 6: Reusable inference rule automating the annotation
of superimposable images.
in the context of several service implementations real-
izing a same kind of treatment. Let us consider the de-
ployment of a new registration service, implemented
with a new algorithm. As soon as this new service
is annotated with Role and Natural concepts of the
same class as (or subsumed by) the concepts appear-
ing in the registration inference rule, there is no need
to rewrite an inference rule specific to this particu-
lar service. As a consequence, workflows involving
this new service will also benefit from the generation
of annotations stating the “superimposability” of data.
With this approach, domain expert can equip their on-
tologies with inference rules that provide meaningful
information to end-users independently from the ser-
vices deployed. Service providers can focus on their
services, transparently reusing such high-level infer-
ence rules.
6 IMPLEMENTATION
6.1 System Architecture
Figure 7 schemes the NeuroLOG platform, with a par-
ticular focus on its semantic components aiming at
enhancing the sharing and enactment of neuroimag-
ing workflows. This deployment shows three col-
laborating sites A, B, C and end-users interacting
with their proper site gateway (Site A) through the
client application. Processing tools are syntactically
described and instrumented as relocatable bundles
through jGASW (Rojas Balderrama et al., 2010) to
enable their deployment and invocation on various
computing infrastructure. The MOTEUR (Glatard
et al., 2008) component enables the design of new
experiments as scientific workflows and is respon-
sible for their enactment. The semantic annotation
of jGASW services is realized through a dedicated
user interface of the client application (Service an-
notator) while the workflow enactor is responsible
for recording provenance information at runtime and
populating the semantic store with OPM RDF state-
ments. Semantic annotations are managed through lo-
cal RDF triple stores implemented with the Jena API.
The CORESE semantic engine (Corby et al., 2004)
is used to perform semantic querying and reasoning
over the knowledge base. CORESE is a semantic
query/rules engine based on conceptual graphs, sup-
porting RDF(S) entailments and a subset of OWL-
Lite entailments: datatypes, transitivity, symmetry
and inverse properties.
Collaborating Site AClient application
Service wrapping
(jGASW)
Workflow designer
and enactor
(MOTEUR)
Semantic query client
and service annotator
Computing and storage infrastructure (EGI grid)
Middleware
(NeuroLOG)
Semantic store
Semantic engine
(CORESE)
Collaborating
Site C
Collaborating
Site B
Middleware
stack
Middleware
stack
Ontology
(OntoNeuroLOG)
Figure 7: NeuroLOG platform: semantic enhancements to
support the sharing and enactment of neuroimaging work-
flows.
6.2 MOTEUR-S
This section presents the extension of the MOTEUR
workflow environment to (i) annotate and catalog
jGASW services and (ii) track and query provenance
information through the OPM standard. This work
is based on the integration of the JSPF plugin frame-
work within MOTEUR, allowing third-party develop-
ers to integrate repository or listener plugins. Reposi-
tory plugins are dedicated to extending the sources of
composable services, and listener plugins (based on
the observer design pattern) are dedicated to monitor
workflow states at runtime, and trigger specific pre-
or post-processing.
6.2.1 Semantically Annotating and Cataloging
Processing Tools
The annotation task, generally performed by the pro-
cessing tool provider, consists in associating to each
service port through a dedicated GUI, at maximum
one Natural concept specifying the semantic nature
of the consumed/produced data, and at maximum one
Role concept characterizing how data is related to the
service through this particular port. It is not desir-
able to associate more than one Natural concept or
CHARACTERIZING SEMANTIC SERVICE PARAMETERS WITH ROLE CONCEPTS TO INFER
DOMAIN-SPECIFIC KNOWLEDGE AT RUNTIME
67
one Role concept to a given service port since it will
conduct to an ambiguous semantic description of the
service. From the service point of view, it would
not be possible to determine which Natural or Role
concept characterizes the consumed or produced data.
Semantic descriptions can then be saved as a collec-
tion of RDF statements, or directly published into the
semantic store.
To enhance the overall coherency of workflows
at design time, the semantic service catalog can be
queried to retrieve services realizing a particular kind
of treatment (through an associated Natural concept
of the domain ontology), or to retrieve services able
to consume a particular kind of data at a given step of
the workflow construction.
6.2.2 Recording and Querying Provenance
Information
When a workflow is started, a new OPM account is
registered and timestamped. One OPM account is cre-
ated per workflow invocation, thus easing the retrieval
of all provenance annotations generated in the context
of a single workflow invocation. For each process
invocation, we register an OPM process entity, also
timestamped, and its consumed and produced data
as OPM artifacts linked to the OPM process through
their corresponding causal dependencies used or was-
GeneratedBy. For each causal dependency, we as-
sociate an OPM role corresponding to the Role con-
cept used to annotate the service description. Finally,
OPM processes are linked through an OPM wasCon-
trolledBy entity to an OPM agent. This agent cor-
responds to the service description identified by the
WSDL URL of the jGASW service deployed. Se-
mantic service description being also identified by the
WSDL URL, the system is thus able to retrieve, from
an OPM process and the semantic catalog of anno-
tated services, all available domain-specific annota-
tions (Natural and Role concepts) describing the in-
voked service.
The CORESE semantic engine is used to per-
form generic queries to retrieve, for instance, from
the workflow results, the source data that has been
derived through the data analysis workflow. In prac-
tice we rely on SPARQL 1.1 property path expres-
sions which provide a compact and powerful language
to handle complex graph matching, such as alterna-
tive/optional paths, or path length constraints.
6.3 New Knowledge Generation
CORESE provides a forward chaining engine that,
from a set of inference rules, saturates its conceptual
graph until no new statements can be inferred. Infer-
ence rules are expressed through the CORESE rule
syntax (non-XML but SPARQL-like), which is very
similar to the SWRL proposal from the W3C, also
describing an implication between an antecedent and
a consequent. However, CORESE rules differ since
they benefit from a limited support of CORESE for
OWL-Lite entailments.
In this first implementation, we assume that infer-
ence rules instrumenting the domain ontology are pro-
vided by the ontology developers and are thus pack-
aged within the domain ontology as complementing
files. Inference rules could be applied to extend the
knowledge base at any time. Rather than letting the
end-user select the suitable inference rule, and trigger
the application of the rule, all available rules are sys-
tematically applied, triggered by the end of a work-
flow invocation through a specific MOTEUR event.
We consider this automatic application of rules be-
cause if the antecedent of the rule is not matched, then
the rule is not applicable. On the other hand, when an
antecedent is matched, it makes sense to apply the
rule since it has been provided by the ontology devel-
opers and has been designed to serve the concerns of
the whole user community.
When the MOTEUR listener plugin dedicated to
provenance annotations is notified with the end of a
workflow invocation, the CORESE semantic engine
is populated with (i) the domain ontology (covering
both Nature and Role concepts) and OPM provenance
ontology, with (ii) all available inference rules pro-
vided with the domain ontology, and (iii) with state-
ments describing the annotated services and OPM
statements describing the workflow invocation. Then
the forward chaining engine of CORESE is started to
produce new inferred statements.
7 CONCLUSIONS AND
PERSPECTIVES
E-Science experimental platforms strongly rely on
Service Oriented Architectures to assemble flows of
data analysis services. However, their usability is
hampered by the level of expertise of experiment de-
signers, as they are expected to have a clear under-
standing of the semantics of the data processing, i.e.
what kind of data is processed and how they are effec-
tively processed. To improve their usability and as-
sist end-users, ontologies, semantic annotations and
reasoning engines are integrated. In this paper, we
proposed a clear delineation between Role and Natu-
ral concepts in the domain ontology to disambiguate
semantic annotation of service parameters. In addi-
KEOD 2011 - International Conference on Knowledge Engineering and Ontology Development
68
tion, the domain ontology can be instrumented with
inference rules that leverage the description of Roles
combined with generic provenance information to en-
rich our semantic repository with meaningful domain-
specific annotations at runtime.
Since this work was implemented in the context
of the NeuroLOG experimental platform, the service
considered for annotations are web services wrapped
with jGASW, a legacy processing tools wrapper tar-
geting large scale distributed infrastructures. How-
ever our approach is more widely applicable, and
could be implemented using standard web services
described through standard service ontologies.
Regarding the sharing of the Role taxonomy of
neuroimaging data with other user communities, two
approaches could be considered as a continuation of
this work: (i) the creation of an OPM profile dedi-
cated to the neuroimaging domain, and (ii) the articu-
lation of the OPM ontology with the DOLCE founda-
tional ontology.
OPM profiles constitute a good opportunity to
share knowledge associated to the role of neuroimag-
ing data. Indeed, an OPM neuroimaging profile could
be constituted with the two subsets of the OntoNeu-
roLOG ontology supporting this work, the Dataset
ontology to extend OPM Artifacts, and the Dataset-
processing ontology to extend OPM Processes. The
Role taxonomy proposed in this paper could be inte-
grated almost directly.
The second approach, more conceptual, would
consist in proposing an OPM ontology whose main
classes are grounded to foundational ontologies such
as DOLCE or BFO (Basic Formal Ontology). It
would allow to smartly articulate OPM and domain
ontologies based on foundational ontologies such as
BIOTOP (Top-Domain ontology for the life sciences)
or OBI (Ontology of Biomedical Investigation) life
science ontologies, and thus exploit these ontologies
at workflow runtime. Indeed considering our ap-
proach from an ontology design perspective, a signifi-
cant effort is still needed for a complete integration in
the OntoNeuroLOG framework, since Role concepts,
designated through the refers-to property, should con-
form to the DOLCE foundational ontology and its re-
lated core ontologies. This ontology integration task
could also cover the semantic overlap between OPM
Artifacts and OntoNeuroLOG Datasets. However, in
the context of this work, the CORESE semantic en-
gine can still (i) retrieve service description, or prove-
nance statements through SPARQL queries and (ii)
produce new meaningful statements through its infer-
ence engine.
The concepts developed in this paper are currently
being integrated in a prototype platform. In the fu-
ture, its use in production in the context of the Virtual
Imaging Platform (VIP project) will enable the eval-
uation of our approach. We plan to study the impact
of Role concepts on four actors in the system: the ser-
vice providers, the workflow designers, the ontology
and inference rule designers, and the final end-users
realizing e-Science workflows. Indeed, we plan to
measure if Role concepts are actually used by service
providers to annotate their processing tools, and if
they enable to disambiguate service parameter anno-
tations, to finally enable more accurate results when
workflow designers query the semantic catalog of ser-
vices. Moreover, we plan to evaluate if Role concepts
are actually involved by ontology designers into in-
ference rules to produce new domain specific state-
ments. Finally, we plan to evaluate the production of
new annotations at workflow runtime, and its useful-
ness from the end-user perspective through the analy-
sis of the semantic queries. More precisely, we want
to determine if the targets of the semantic queries are
annotations inferred from rules involving roles, or if
the targets are annotations produced by other means.
Initially applied to computational neurosciences,
this work goes beyond this scope, as same principles
are planned to be applied in the context of the VIP
project, which targets medical image acquisition sim-
ulation. It is envisaged to validate the applicability
and usability of the delineation of Role and Natural
concepts in domain ontologies to (i) ease the design of
simulation workflows (e.g. simulated cardiac images
through ultrasound modality) and (ii) extend semantic
repositories with new meaningful statements describ-
ing either simulated data or the simulated organs and
their constituting anatomical entities.
ACKNOWLEDGEMENTS
This work was funded by the French National Agency
for Research under grants ANR-06-TLOG-024 (Neu-
roLOG project) and ANR-09-COSI-03 (VIP project).
REFERENCES
Battle, S., Bernstein, A., Boley, H., Grosof, B.,
Gruninger, M., Hull, R., Kifer, M., Martin, D.,
McIlraith, S., McGuinness, D., Su, J., and Tabet,
S. (2005). Semantic Web Services Ontology
(SWSO) [http://www.w3.org/submission/swsf-swso].
http://www.w3.org/Submission/SWSF-SWSO.
Bhagat, J., Tanoh, F., Nzuobontane, E., Laurent, T., Or-
lowski, J., Roos, M., Wolstencroft, K., Aleksejevs, S.,
Stevens, R., Pettifer, S., Lopez, R., and Goble, C. A.
(2010). BioCatalogue: a universal catalogue of web
CHARACTERIZING SEMANTIC SERVICE PARAMETERS WITH ROLE CONCEPTS TO INFER
DOMAIN-SPECIFIC KNOWLEDGE AT RUNTIME
69
services for the life sciences. Nucleic Acid Research,
Article in press.
Corby, O., Dieng-Kuntz, R., and Faron-Zucker, C. (2004).
Querying the Semantic Web with Corese Search En-
gine. In ECAI, pages 705–709.
Glatard, T., Montagnat, J., Lingrand, D., and Pennec, X.
(2008). Flexible and efficient workflow deploye-
ment of data-intensive applications on grids with MO-
TEUR. International Journal of High Performance
Computing Applications (IJHPCA) IF=1.109 Special
issue on Special Issue on Workflows Systems in Grid
Environments, 22(3):347–360.
Gordon, P. M. K. and Sensen, C. W. (2008). Creating Bioin-
formatics Semantic Web Services from Existing Web
Services: A Real-World Application of SAWSDL. In
ICWS, pages 608–614.
Gruninger, M., Hull, R., and McIlraith, S. (2008). A Short
Overview of FLOWS: A First-Order Logic Ontology
of Web Services. IEEE Data Engineering Bulletin.,
31(3):3–7.
Henriksson, J., Pradel, M., Zschaler, S., and Pan, J. Z.
(2008). Ontology Design and Reuse with Conceptual
Roles. In Proceedings of the 2nd International Con-
ference on Web Reasoning and Rule Systems, RR ’08,
pages 104–118, Berlin, Heidelberg. Springer-Verlag.
Kassel, G. (2005). Integration of the DOLCE top-level
ontology into the OntoSpec methodology. Technical
Report 2005-10-18, University of Picardie, Amiens,
France.
Kopeck
´
y, J., Vitvar, T., Bournez, C., and Farrell, J. (2007).
SAWSDL: Semantic Annotations for WSDL and
XML Schema. IEEE Internet Computing, 11(6):60–
67.
Lord, P., Alper, P., Wroe, C., and Goble, C. (2005). Feta:
A Light-Weight Architecture for User Oriented Se-
mantic Service Discovery. In European Semantic Web
Conference, pages 17–31. Springer Berlin / Heidel-
berg.
Martin, D., Burstein, M., McDermott, D., McIlraith, S.,
Paolucci, M., Sycara, K., McGuinness, D. L., Sirin,
E., and Srinivasan, N. (2007). Bringing Semantics
to Web Services with OWL-S. World Wide Web,
10(3):243–277.
Montagnat, J., Gaignard, A., Lingrand, D., Rojas Balder-
rama, J., Collet, P., and Lahire, P. (2008). NeuroLOG:
a community-driven middleware design. In Health-
Grid, pages 49–58, Chicago. IOS Press.
Moreau, L., Clifford, B., Freire, J., Futrelle, J., Gil, Y.,
Groth, P., Kwasnikowska, N., Miles, S., Missier, P.,
Myers, J., Plale, B., Simmhan, Y., Stephan, E., and
den Bussche, J. V. (2011). The Open Provenance
Model core specification (v1.1). Future Generation
Computer Systems, 27(6):743 – 756.
Rojas Balderrama, J., Montagnat, J., and Lingrand, D.
(2010). jGASW: A Service-Oriented Framework
Supporting High Throughput Computing and Non-
functional Concerns. In IEEE International Confer-
ence on Web Services(ICWS 2010), Miami, FL, USA.
IEEE Computer Society.
Roman, D., de Bruijn, J., Mocan, A., Lausen, H.,
Domingue, J., Bussler, C., and Fensel, D. (2006).
WWW: WSMO, WSML, and WSMX in a Nutshell.
The Semantic Web – ASWC 2006, pages 516–522.
Sheth, A. P., Gomadam, K., and Ranabahu, A. (2008). Se-
mantics enhanced Services: METEOR-S, SAWSDL
and SA-REST. IEEE Data Eng. Bull., 31(3):8–12.
Sowa, J. F. (1984). Conceptual structures: information pro-
cessing in mind and machine. Addison-Wesley Long-
man Publishing Co., Inc., Boston, MA, USA.
Temal, L., Dojat, M., Kassel, G., and Gibaud, B. (2008). To-
wards an ontology for sharing medical images and re-
gions of interest in neuroimaging. Journal of Biomed-
ical Informatics, 41(5):766–778.
Vitvar, T., Kopecky, J., Viskova, J., and Fensel, D. (2008).
WSMO-Lite Annotations for Web Services. In 5th Eu-
ropean Semantic Web Conference (ESWC2008), pages
674–689.
Withers, D., Kawas, E., McCarthy, L., Vandervalk, B.,
and Wilkinson, M. (2010). Semantically-guided
workflow construction in Taverna: the SADI and
BioMoby plug-ins. In Proceedings of the 4th interna-
tional conference on Leveraging applications of for-
mal methods, verification, and validation - Volume
Part I, ISoLA’10, pages 301–312, Berlin, Heidelberg.
Springer-Verlag.
Wolstencroft, K., Alper, P., Hull, D., Wroe, C., Lord, P. W.,
Stevens, R. D., and Goble, C. A. (2007). The
my
Grid
ontology: bioinformatics service discovery. IJBRA,
3(3):303–325.
KEOD 2011 - International Conference on Knowledge Engineering and Ontology Development
70