A recent trend in a number of academic disciplines is the publication of results of experiments together with the scienti c article for a better reproducibility of the published experiments and algorithms. Semantic Web technologies have the potential to aid scientists in the publishing, sharing and interlinking of this data and also in helping other scientists in the understanding of the data and the interpretation of the results of an experiment. In this paper we report on a use case on how to publish the data captured in a scienti c experiment that has been conducted in the CSIRO Animal, Food and Health Sciences division as a set of ontologies and how to access this data through a set of RESTful semantic Web services. These services showcase how computational tasks that cannot be represented in the ontology can be implemented as lightweight semantic Web services to document and verify the results of an experiment. Together, the ontologies, the experimental data and the computational services constitute the elements needed for a semantically enabled lab notebook, facilitating research studies over multiple experiments, while reducing complexity and error rates.
Research into di erent types of thermal treatments as quarantine methods against codling moth in a variety of fruit has gained much interest in recent years due to the uncertain future of chemical fumigation of food and the public concern over residues in treated products [25]. One such new thermal treatment method is currently under development in the CSIRO Animal, Food and Health Sciences division. The method proposes to use a combination of thermal treatment with microwave treatment of fruit to inactivate the fruit y larvae from growth in di erent types of fruit. The experiment conducted by domain scientists in CSIRO is using a custombuilt microwave-based heat treatment system that is tested on a number of di erent fruit for the inactivation of an induced codling moth infestation in these fruit.
Apart from the main goal of the research in establishing the e ectiveness of the microwave-based heat treatment process for the inactivation of codling moth, a secondary goal of a transformational capability platform project was to showcase how the experimental data can be modelled in semantic Web languages and how the resulting ontological models can be used to support computational analysis of the experimental data. To achieve the latter, a group of ontologists and software engineers have accompanied the domain scientists during the experiment and de ned models to capture the experimental data semantically. We present our methodology of how to publish the results of a scienti c experiment as a set of ontologies. Further, we develop a set of services that showcase how computational tasks that cannot be represented in the ontologies themselves can be modelled and implemented as lightweight semantic Web services.
The remainder of this paper is structured as follows. In Sect. 2 we brie y describe the microwave-based heat treatment process that constitutes our use case. In Sect. 3 we describe the ontologies that are needed to capture this use case. In Sect. 4 we describe the information services that we have built on top of the ontological data representing the knowledge gathered in the use case experiment. We discuss some related work in Sect. 5, before we conclude in Sect. 6. 2
Our use case has been provided by scientists in the CSIRO Animal, Food and Health Sciences (CAFHS) who are developing a microwave-based heat treatment system for the treatment of fruit for the purposes of removing fruit y larvae infestations. Brie y, for the experiment a purpose-built microwave tunnel system was built. The microwave unit also incorporates an auxiliary hot air system comprised of a heater and a fan which is also attached to the microwave system.
For the experiment that we modelled ontologically, newly harvested organically grown Mutsu and Granny Smith apples were used for treatment in this system. The apples were uniformly infested with fruit ies by making 50 pin holes on each apple at the stem end and then placed inside cages containing fruit ies.
The microwave treatment was applied by placing the apples on a small plastic stand with four protruding rods and sent through the microwave tunnel for approximately 54 min, which was preheated to 63 { 65 C. A variable speed conveyor belt moves the fruit through the microwave tunnel where they undergo heating by microwaving to destroy fruit y larvae and eggs. The temperature of the fruits at di erent points (top, esh, core, bottom) were measured during the experiments with a bre optic conditioner at 1 second intervals.
The goal of the experiment was to determine the optimal con guration of the tunnel temperature, the microwaving intensity and the time of treatment in each of the stages of the treatment process to obtain 100% mortality of the fruit y larvae and eggs that is comparable to traditional thermal treatment methods. This can be done by calculating the cumulative thermal e ect for a given treatment, based on kinetic data for the thermal mortality of target insects and for product quality losses. Given a time-temperature history of T (t), the cumulative thermal mortality of the microwave-based heat treatment can be calculated to an equivalent length of time in minutes, M52, at a reference temperature T ref of 52 C by using the following relationship: M52 =
Z t 0 10
T (t) z52 C
dt where M52 is the equivalent time at a target temperature of 52 C, T (t) is the transient temperature pro le measured by the bre optic system, t is the time and z is the temperature change (in C) required to change the value of insect mortality (lethality) by a factor of 10.
To unambiguously record the data (e.g. temperature measurements, applied power, belt speed etc.) captured in the experiment and to allow a computational analysis of the experimental data, we rst need a conceptual strati cation of the experiment and a common understanding of the objects and processes that are used in the experiment. In the context of this project, a fruit treatment process use-case acts as an exemplar to build up a demonstrator of a semantics driven lab notebook.
When describing experiments in electronic lab notebooks the terms used are often ambiguous. In our use case experiment, for example, the term \fruit" if used in some electronic record, may be ambiguous depending on the experiment run, as there were multiple experiments conducted, not only with Mutsu and Granny Smith apples, but also with mangoes and avocados. Even more, many of these terms exhibit polysemy: avocado, in common usage - and therefore also when used as a metadata term - may, without further speci cation, refer to either the \avocado fruit" or the \avocado tree". Such distinctions are important in that they (a) determine the scope of what we can talk about in our information systems and (b) also specify - at least to a degree what sort of data is to be collected. Completely de ning the meaning of something allows the speci cation of the relationships between entities: an \avocado fruit", for example, is part of an \avocado tree" (at least until it has been harvested). Such a disambiguation will then, for example, allow us to talk about properties of a speci c fruit (e.g. volume, rmness) and its history and provenance (this \avocado fruit" was part of an \avocado tree" which was located in a \ eld" which is described by \geo-coordinates" X and Y etc.). Without disambiguating the polysemous term \avocado" it would have been impossible to represent information about an avocado in such terms.
The precise de nition of objects and their relationships can also help to overcome the strati cation in conceptual models of the treatment system, i.e. its factory: an apple has a digital representation denoting the apple in an information system, i.e. in the electronic lab notebook { for the purposes of this description we will call it a \digital apple". The \digital apple" is described by some \apple description", which, in turn is a kind of \information content entity". An \information content entity", in turn, may be an input into a \model", for example, a model describing the relationship between the apple volume and the required heating intensity to reach a certain core temperature in the apple.
In summary, a semantic lab notebook is an exercise in object management. For the purposes of the rest of this discussion, the term \object" denotes any entity that can be named or addressed. Objects may therefore be physical objects as well as data objects, computational service objects etc..
We have chosen to model these objects with the languages developed in the
technology stack of the semantic Web [
To develop the ontology for the use case outlined above, we used a modi ed and
shortened version of the method described by Uschold and King [26]. For the
purposes of ontology development we used documentation containing domain-speci c
terminology and data as elicited from our colleagues at CAFHS. Fig. 1 shows the
stack of ontologies we reused and developed within this project. The gure also
includes example classes that are de ned within each of these ontologies, whereas
the dashes denote the subsumption relations between the classes. In the following
sections we describe the classes and relations in these ontologies in more detail.
Upper Level Ontology { GFO Upper{ or \top level" or \foundational" {ontologies
are ontologies of the most common entities in the world which are the same across all
knowledge domains. For example, an upper ontology will provide a notion of what a
material entity is, how material entities participate in processes and persist in time.
The main purpose of an upper ontology is to facilitate semantic interoperability. A
number of upper level ontologies are in use across the semantic Web community,
though many Web ontologies are developed without referencing top level ontologies,
often reducing the level of interoperability. Some of the most common ontologies
currently in use across the semantic Web are DOLCE [18], the General Formal Ontology
(GFO) [
Science Ontology The Science Ontology [
Information Content Entities The most relevant concept that we reuse from the
Science Ontology is that of an \Information Content Entity" (ICE) that is required
to capture data, speci cations and descriptions. ICEs are best described as entities
that do not have independent existence, but rather are dependent on other entities
and are in an \about-ness" relationship with those entities [
While perdurants such as processes may map onto a time vector, the entity that we observe when we measure time, for example, is not the time vector itself, but a representation, or in better terms, a descriptions of the time vector. In our ontology, processes therefore have \process speci cations" which are in an aboutness relationship to the process itself { process speci cations are types (subclasses) of descriptions, which, in turn, are information content entities. We may write:
Processing Ontology The microwave-based heat treatment process conducted in our use case is some kind of a food treatment process. Consequently, we need an ontology that de nes concepts and relations of food treatment processes. However, to the best of our knowledge no such food treatment ontology exists. De ning such an ontology requires us to properly layer it on top of an Upper Level Ontology such as GFO. To do that we need to identify what constitutes a food treatment process and what are the more general concepts and relations that are needed to describe such a process. Looking at the concept of a food treatment process, it obviously involves some kind of \treatment process" that is performed on \food". A \treatment process" is performed either by a \human" or a \machine". The GFO makes a fundamental distinction between \Processes" and \Actions": ontologically, both are viewed as being of type \Occurrent" but are distinguished from each other through the involvement of an \Agent", i.e. an entity, playing an agent role. Agent roles can be played by both humans and machines. The entity that is treated during the treatment process is at its most general a gf o:M aterial object. \Food" is { ontologically speaking { a role: a material object \becomes food" when it realises the food role (an apple sitting on a shelf, for example, is not \food" as it does not realise the food role).
From this very brief ontological analysis it becomes clear that we rst need an
ontology that describes objects in a factory, the roles these objects play as well as the
processes in a factory and the mode of participation in those processes. A number
of such manufacturing/processing ontologies exist already, however, they are either
not layered on top of an Upper Level Ontology [
Processes in the processing ontology are layered on the notion of processes in GFO which are characterised by the manner in which entities participate in them, i.e.:
We de ned a set of such roles that are common in manufacturing processes and that are played by material objects, such as a \heated entity role" and a \heating entity role". We may write:
Further, we de ned \chemical material objects", \physical material objects" and \biological material objects" that manufacturing processes take as input or produce as output such as a \machine", an \assembly entity" and di erent types of \substances".
Processes can have other processes as part which allows the modeling of complex
manufacturing processes and their breakdown into small process parts. Any other
process can be modeled by analogy. Processes in our ontology have time
boundaries with discrete start and end timepoints. The time boundaries for processes
are mapped to the notion of \Chronoids" in GFO. Time is understood in GFO to
be Brentano time [
This provides all the mechanisms needed to de ne process durations as well as start and end times and dates.
Food Processing Ontology On top of the manufacturing processing ontology we have developed a generic food processing ontology, denoted by the namespace pre x \fp" in Fig. 1 and a more use case-speci c apple processing ontology, denoted by the namespace pre x \ap". To the best of our knowledge, there exist no such ontologies, but for an improved interoperability we have included equivalence relations to the NCI Thesaurus1 for all the biological concepts that are de ned in the food processing ontology and apple processing ontology.
Material Objects Much of the use case experiment is concerned with apple processing and hence, apples can serve as an illustration of how we handle material objects in the ontology. As discussed above, the term \apple" is potentially polysemous and hence, we need to distinguish between an \apple tree" and an \apple fruit". Furthermore, \apple fruit" must be subdivided into several types of apples such as \apple fruit on tree", \harvested apple fruit" or \refrigerated apple fruit" if we wish to talk about fruit still ripening on trees as opposed to harvested ones and ones which have undergone some treatment. Ontologically speaking, all of these entities are subclasses of the GFO's \material object" class. We may therefore write in First Order Logic: 1. ap:f ruit subclassOf ap:organism part
Processes The ontological treatment of processes in the food processing ontology is analogous to the description outlined above: the apple microwave treatment process has at least three discernible participants and distinct roles: (a) an apple playing the role of the heated entity, (b) the microwave oven playing the role of the treating 1 http://ncit.nci.nih.gov/ entity and (c) an apple playing the role of the treated entity. The apple in (a) and (c) are ontologically distinct: with the beginning of treatment process, the apple has ceased to be an untreated apple and become a treated apple. In rst order description logic, we may de ne the treatment process as follows: 1. ap:apple treatment process equivalentClassOf (f p:microwave heating process and (proc:has participant some (ap:apple f ruit and (gf o:plays role some proc:heated entity role))) and (proc:has participant some (proc:microwave oven and (gf o:plays role some proc:heating entity role))) and (has participant some (apple f ruit and (gf o:plays role some proc:treated entity role))))
f p:microwave heating process
4
For our proof-of-concept service implementation to analyse and verify the
experimental data we have chosen to use the SADI{semantic Automated Discovery and
Integration{framework [27]. SADI is a semantic Web service framework that is
predominantly used in the bioinformatics domain. SADI comprises a set of semantic
Web compliant conventions and suggested best-practices for data representation and
exchange between Web services. In contrast to other semantic Web service
frameworks, SADI takes some assumptions that make the protocol and the implementation
much easier than, for example, OWL-S [17] and WSMO [22]/WSMX [
We deployed our SADI information services onto an Apache Tomcat Server and use the Jena library to query the RDF Triple Store running on the same Tomcat instance (see Fig. 2). The RDF database is loaded with the ontologies as described in Sect. 3.1 constituting the TBox, while the actual data (the ABox ), such as the temperature measurements for apples undergoing the heat treatment from di erent runs of the experiment, are loaded into the Triple store via scripts that transform the raw sensor data into ontology instances. The services use SPARQL queries to retrieve the required information from the Triple store, process them and return the annotated ontology instance back to the client. The rst semantic Web service we developed for analysing data in the semantic lab notebook allows to verify if an apple undergoing the treatment process was refrigerated properly before the experiment. The service takes as input (see Listing 1.1) a speci c refrigeration process speci cation and gives a boolean return (see Listing 1.2) con rming if all the temperature observations recorded for the apples participating in the refrigeration process comply to the limits de ned in the speci cation. The refrigeration process itself references the individual apples/batches through a gf o:plays role relation. The semantic Web service uses SPARQL to query the speci cation of the provided process and to extract the allowed minimum and maximum temperature values for each refrigeration process. Then all temperature observations for apples playing a role in the given process are extracted and checked against the speci ed limits. A boolean attribute f p:isCertif iedP rocess is then added to the instance of the process speci cation which is in turn returned by the service.
A false return value would indicate an interrupted cooling chain and thus nullify the results of the experiment.
Listing 1.1. Input RDF <!DOCTYPE r d f :RDF
[ <!ENTITY m a t i n f " http : / / m a t i n f . cmse . c s i r o . au/"> ]> <r d f :RDF xmlns : r d f ="http : / /www. w3 . org /1999/02/22 r d f syntax ns#" xml : base="@matinf ; i d / exp 12 09 12#" xmlns : f p="@matinf ; ont / owl / p r o c e s s i n g . owl#""> <owl : Ontology r d f : about="@matinf ; i d / exp 12 09 12#"/> <f p : r e f r i g e r a t i o n p r o c e s s s p e c i f i c a t i o n
r d f : about="@matinf ; i d / exp 12 09 12# r e f r i g e r a t i o n p r o c s p e c 1 "> </f p : r e f r i g e r a t i o n p r o c e s s s p e c > </ r d f :RDF>
Listing 1.2. Output RDF <!DOCTYPE r d f :RDF
[ <!ENTITY m a t i n f " http : / / m a t i n f . cmse . c s i r o . au/"> ]> <r d f :RDF xmlns : r d f ="http : / /www. w3 . org /1999/02/22 r d f syntax ns#"
This service essentially implements a complex SPARQL query using FILTERS
and thus could also be expressed in a SPARQL templating language such as SPIN [
The second service expects an instance of an \apple description" as input (see Listing 1.3) and returns a M52 time equivalent [25] (e.g. \m52 model output speci cation 1" in Listing 1.4) that has been achieved during a microwave treatment process for the given apple (described by the apple description). The returned \m52 model output speci cation 1" instance is a speci cation itself while the actual value that was created for the M52 time equivalent can be queried via the following SPARQL query: SELECT ?o WHERE { <base:m52_model_output_specification_1> <science:has_Value_Literal> ?o }
In our case this query returns a value of 22.62 minutes which indicates how long the embedded larvae in the speci c apple would have been exposed to a reference temperature of 52 C in the treatment process. To calculate this value, the service retrieves all temperature observations for each of the four sensors embedded in the speci c apple described by \granny smith apple desc 001" and incrementally accumulates the minimum accumulated total temperature equivalent to M52. This value indicates if the apple was su ciently heat treated in the experiment to kill all embedded larvae.
Listing 1.3. Input RDF <!DOCTYPE r d f :RDF
[ <!ENTITY m a t i n f " http : / / m a t i n f . cmse . c s i r o . au/"> ]> <r d f :RDF xmlns : r d f ="http : / /www. w3 . org /1999/02/22 r d f syntax ns#" xml : base="@matinf ; i d / exp 12 09 12#" xmlns : ap="@matinf ; ont / owl / a p p l e p r o c e s s i n g . owl#"> <owl : Ontology r d f : about="@matinf ; i d / exp 2012 09 12#"/> <ap : a p p l e d e s c r i p t i o n
r d f : about="@matinf ; i d / exp 12 09 12# g r a n n y s m i t h a p p l e d e s c 0 0 1 "> </ap : a p p l e d e s c r i p t i o n > </ r d f :RDF>
Listing 1.4. Output RDF
This is not the rst time that semantically enabled electronic lab notebooks have
been proposed [
Some others have developed tools focussed on di erent aspects of the experimental data curation problem, for example on the scalability [23], or on capturing the relationships between results of di erent experiments [19].
[20] proposes an aggregation tool based on RSS feeds to ensure that the objects created during the research process are recognized, stored and indexed.
The bioinformatics community as a whole is spearheading other academic disciplines by capturing vast quantities of the knowledge published in scienti c articles as ontologies in the Bioportal initiative2.
However, we are not aware of any prior works on capturing the data produced by an experiment in ontologies combined with custom-built RESTful semantic Web services on top of the RDF data that allow to reproduce and verify the results of the experiment. The closest work to ours, but with a stronger focus on capturing the entire work ow of an experiment was proposed in [16]. The work introduces a laboratory domain speci c ontology and the COW (Combining Ontologies with Work ows) software tool was developed to formalize work ows which were enhanced with ontological concepts taken from the developed domain speci c ontology. 6
We described our prototypical implementation of a semantic lab notebook that allows data obtained in an experiment to be stored in RDF and accessed via SPARQL and custom-built semantic Web services. The system allows scientists to read the experiment related data and to combine it as part of a scienti c work ow. We implemented ontologies required to model our data points via Protege in OWL and 2 http://bioportal.bioontology.org/ developed SADI RESTful Web services in Java that implement computational analysis functionality that cannot be expressed in the ontology directly. With an increased availability of ontologies and tools that support the capture of RDF, semantic lab notebooks can play a signi cant role in helping the research community to store experiment data consistently, process it faster and allow the mashup of collected datasets to facilitate research studies over multiple datasets, while reducing complexity and error rates.
Acknowledgements We are thankful to the support of our colleagues in the CSIRO Animal, Food and Health Sciences division, in particular to Dr. Mala Gamage who regularly shared the results of their heat-treatment experiments with us. 15. H. Lin and J. Harding. A manufacturing system engineering ontology model on the semantic web for inter-enterprise collaboration. Computers in Industry, 58(5):428{437, 2007. 16. A. Maccagnan, M. Riva, E. Feltrin, B. Simionati, T. Vardanega, G. Valle, and N. Cannata. Combining ontologies and work ows to design formal protocols for biological laboratories. Automated Experimentation, 2, 2010. 17. D. Martin, M. Burstein, J. Hobbs, O. Lassila, D. McDermott, S. McIlraith, S. Narayanan, M. Paolucci, B. Parsia, T. Payne, E. Sirin, N. Srinivasan, and K. Sycara.
OWL-S: Semantic Markup for Web Services. Member submission, W3C, 2004. 18. C. Masolo, S. Borgo, A. Gangemi, N. Guarino, A. Oltramari, R. Oltramari, L. Schneider, L. P. Istc-cnr, and I. Horrocks. Wonderweb deliverable d17. The Wonderweb library of foundational ontologies and the DOLCE ontology. 2002. 19. J. Myers, C. Pancerella, C. Lansing, K. Schuchardt, and B. Didier. Multi-Scale Science, Supporting Emerging Practice with Semantically Derived Provenance. In ISWC workshop on Semantic Web Technologies for Searching and Retrieving Scienti c Data, 2003. 20. C. Neylon. Head in the clouds: Re-imagining the experimental laboratory record for the web-based networked world. Automated Experimentation, 1, 2009. 21. E. Polonsky, E. Polonsky, A. Six, M. Kotelnikov, V. Polonsky, R. Polly, and P. Brey.
Semantic laboratory notebook: Managing biomedical research notes and experimental data. In Proceedings of ESWC, 2006. 22. D. Roman, U. Keller, H. Lausen, J. de Bruijn, R. Lara, M. Stollberg, A. Polleres, C. Feier, C. Bussler, and D. Fensel. Web Service Modeling Ontology. Applied Ontologies, 1(1):77{106, 2005. 23. M. Stonebraker, D. Bruckner, I. F. Ilyas, G. Beskales, M. Cherniack, S. B. Zdonik,
A. Pagan, and S. Xu. Data curation at scale: The data tamer system. In CIDR, 2013. 24. T. Talbott, M. Peterson, J. Schwidder, and J. D. Myers. Adapting the electronic laboratory notebook for the semantic era. In Proceedings of the 2005 international conference on Collaborative technologies and systems, CTS'05, pages 136{143, Washington, DC, USA, 2005. IEEE Computer Society. 25. J. Tang, J. Ikediala, S. Wang, J. Hansen, and R. Cavalieri. High-temperature-shorttime thermal quarantine methods. Postharvest Biology and Technology, 21(1):129 { 145, 2000. 26. M. Uschold and M. King. Towards a Methodology for Building Ontologies. In Workshop on Basic Ontological Issues in Knowledge Sharing, held in conjunction with IJCAI-95, 1995. 27. M. Wilkinson, B. Vandervalk, and M. L. The Semantic Automated Discovery and Integration (SADI) Web service Design-Pattern, API and Reference Implementation. Journal Biomed Semantics, 2, 2011.