Work ow-Centric Research Ob jects: First Class Citizens in Scholarly Discourse

Abstract

A work ow-centric research object bundles a work ow, the provenance of the results obtained by its enactment, other digital objects that are relevant for the experiment (papers, datasets, etc.), and annotations that semantically describe all these objects. In this paper, we propose a model to specify work ow-centric research objects, and show how the model can be grounded using semantic technologies and existing vocabularies, in particular the Object Reuse and Exchange (ORE) model and the Annotation Ontology (AO). We describe the life-cycle of a research object, which resembles the life-cycle of a scienti c experiment.

-

Scienti c work ows are used to describe series of structured activities and computations that arise in scienti c problem-solving, providing scientists from virtually any discipline with a means to specify and enact their experiments [ 3 ]. From a computational perspective, such experiments (work ows) can be de ned as directed acyclic graphs where the nodes correspond to analysis operations, which can be supplied locally or by third party web services, and where the edges specify the ow of data between those operations.

Besides being useful to describe and execute computations, work ows also allow encoding of scienti c methods and know-how. Hence they are valuable objects from a scholarly point of view, for several reasons: (i) to allow assessment of the reproducability of results; (ii) to be reused by the same or by a di erent scientist; (iii) to be repurposed for other goals than those for which it was originally built; (iv) to validate the method that led to a new scienti c insight; (v) to serve as live-tutorials, exposing how to take advantage of existing data infrastructure, etc. This follows a trend that can be observed in disciplines such as Biology and Astronomy, with other types of objects, such as databases, increasingly becoming part of the research outcomes of an individual or a group, and hence also being shared, cited, reused, versioned, etc. [ 11 ]

However, the use of work ow speci cations on their own does not guarantee to support reusability, shareability, reproducibility, or better understanding of scienti c methods. Work ow environment tools evolve across the years, or they may even disappear. The services and tools used by the work ow may change or evolve too. Finally, the data used by the work ow may be updated or no longer available. To overcome these issues, additional information may be needed. This includes annotations to describe the operations performed by the work ow; annotations to provide details like authors, versions, citations, etc.; links to other resources, such as the provenance of the results obtained by executing the work ow, datasets used as input, etc.. Such additional annotations enable a comprehensive view of the experiment, and encourage inspection of the di erent elements of that experiment, providing the scientist with a picture of the strengths and weaknesses of the digital experiment in relation to decay, adaptability, stability, etc.

These richly annotation objects are what we call work ow-centric research objects. The notion of Research Object has been introduced in previous work [ 20, 19, 1 ] { here we focus on Research Objects that encapsulate scienti c workows (hence work ow-centric). In particular, we build on earlier work on myExperiment packs, which are bundles that contain elements such as work ows, documents and presentations [ 15 ]. Other related work is presented in Section 2. In this paper we extend that work making the following contributions: we present a model for specifying work ow-centric research objects (Section 3), and show how it is grounded using semantic technologies; and we characterise and de ne their lifecycle, illustrating how they evolve over time to be augmented with provenance of the work ow results and semantic annotations (Section 4). 2

Related Work

In certain disciplines (e.g., life sciences), scienti c communication channels like journals encourage or mandate authors of submitted papers to include information about the methods used to reach the conclusions claimed in the paper. This has the aim of promoting reproducibility and reuse of the scienti c results reported on those papers. For example, most 'wet lab' life science journal papers must contain a `materials and methods' section that describes the details about the experiments that the authors conducted. These journals typically have strict rules about how to formulate these sections, but from a computational point of view it is weakly structured; hence they are still hard for other scientists to discover and reuse.

The practice of conveying computational methods in a standardised and highly structured way has had less time to evolve in many areas of science. Some journals are also encouraging authors to make available the data and software that have been used and produced, that is, to make data and processes used part of the published work [ 8 ]. For example, Bioinformatics1 considers software availability as an important prerequisite to the acceptance of the paper. And the NASA ADS (Astrophysics Data System)2 is linking and referencing papers, references to the journal, data behind the plots used in the papers, catalogues of objects used (as URL references), software used (as URL references to the Astrophysics Source Code Library), instrument used to gather the observed/input data, and the proposal submitted to ask for observation time. These are important steps forward to promote sharing and reuse. However, software and data availability may not be su cient to check the reproducibility of results, as described in the introduction.

As stated in the introduction, our model is built on earlier work on myExperiment packs [ 15 ], which aggregate elements such as work ows, documents and datasets together, following Web 2.0 and Linked Data principles [ 18, 17 ]. The myExperiment ontology [ 14 ], which forms the basis for our research object model, has been designed such that it can be easily aligned with existing ontologies. For instance, their elements can be assigned annotations comparable to those de ned by Open Annotation Collaboration (OAC).

One important aspect of our work is that we make use of abstract workow templates as a means to annotate work ow templates, facilitating work ow speci cation (as done by Gil et al. [ 6 ] and Ludascher et al. [ 9 ]). Scientists describe a work ow by identifying abstract tasks and specifying scienti c analyses using semantic concepts from an underlying domain ontology. The speci ed abstract work ow is then mapped to a concrete work ow using mappings that specify for each task the underlying service operations that can be used for its implementations.

Our work is complementary to the above proposals in the sense that, in addition to semantic annotations of work ows, we exploit provenance of work ow results to describe work ow templates. In this context, similar proposals are CrowdLab [ 10 ], which provides users with the means for publishing data as well as work ows and the provenance of their results to promote the reproducibility of such results, Janus [ 12 ] and OPMW [ 5 ]. Here we leverage semantic technologies and underline the importance of annotations, which we hope will yield a wide adoption of research objects among scientists. Besides, we allow connecting more elements to the work ow: alternative material, alternative web services, bibliography, the proposal that led to the work ow/experiment, etc.

A clear demand from domains such as bioinformatics and astronomy is the ability to understand a work ow, for which elements outside of the work ow are often needed. 3

A Model for Work ow-Centric Research Objects

Our work ow-centric research object model aims at providing support for the description of the scienti c processes described in the previous section in a machine 1 http://bioinformatics.oxfordjournals.org/ 2 http://labs.adsabs.harvard.edu/ processable format, together with the datasets involved, the results obtained, and their provenance information. The research object will be also accompanied with annotations, which will promote the discover-ability, and therefore the reusability of the processes (work ows), as well as enabling third parties to assess the validity and reproducibility of the results.

Figure 1 illustrates a coarse-grained view of a work ow-centric research object, which aggregates a number of resources, namely: { a work ow template, which de nes the work ow; { work ow runs obtained by enacting the work ow template { other artifacts which can be of di erent kinds, e.g., a paper that describes the research, datasets used in the experiments, etc.; { annotations describing the aforementioned elements and their relationships. Fig. 1: Work ow-centric research object as an aggregation of resources. CHECK IF IT IS OK

Figure 2 provides a more detailed view of the resources that compose workow templates and work ow runs. A work ow template is a graph in which the nodes are processes and the edges represent data links that connect the output of a given process to the input of another process, specifying that the artifacts produced by the former are used to feed the latter. A process is used to describe a class of actions that when enacted give rise to process runs. The process speci es the software component (e.g., web service) responsible for undertaking the action. Note that some work ow systems may specify in addition to the data ow, the control ow, which speci es temporal dependencies and conditional ows between processes. We chose to con ne the work ow research object model to data-driven work ows, as in Taverna [ 16 ], Triana [ 2 ], the process run Network Director supplied by Kepler [ 4 ], Galaxy3, Wings [ 7 ], etc.

Figure 3-b illustrates an example of a work ow template that is composed of two processes. Such a work ow describes an in-silico bioinformatics experiment that is used to identify gene pathways. Speci cally, the work ow is composed of two processes: given a protein accession, the GetKeggGeneId process is used to 3 http://galaxy.psu.edu/ retrieve the corresponding gene ID. The gene ID retrieved is then used to feed the GetKeggPathway process, which returns the corresponding pathways. Note that we also support work ow instances, which are work ow templates with the inputs bound to data values. We also distinguish between standard input parameters and con guration input parameters. Con guration input parameters are used to set the algorithm, the underlying sources used by the processes that compose a work ow template and so on. In addition, the processes that compose a work ow template are not always bound to a software component, rather they can be performed manually in which case they are associated with a human agent.

A work ow template can be instantiated and enacted using a work ow engine, e.g., Taverna. This gives rise to a work ow run that speci es the process runs that were obtained by executing the processes that constitute the workow template in question. For example, when the action speci ed by the process is undertaken by a web service, the process run obtained by enacting such a process represents a web service call. A process run may take as input some existing artifacts, speci ed by the used association, and output some new artifacts, speci ed by the wasGeneratedBy association. Artifact is a general concept that represents an immutable piece of state, which may have a physical embodiment in a physical object, or a digital representation in a computer system [ 13 ]. In the context of work ow-centric research objects, the focus is on artifacts that are digital representations in a computer system. It is worth mentioning that the notion of process run and artifact that we use are aligned with major provenance models such as the Open Provenance Model (OPM) [ 13 ] and PROV-DM4.

Figure 3-c illustrates an example of a work ow run that is obtained by enacting the work ow template together with the provenance of the results produced by the work ow run, which are depicted in Figure 3-b. Get4 http://www.w3.org/TR/2011/WD-prov-dm-20111018

GeneIdRun, and GetGenePathwayRun are process runs that were obtained by enacting the GetGeneId and GetGenePathway processes, respectively. GetGeneIdRun took as input the protein accession up:11005 and generated the gene id syf:Synpcc7942 0655, the process run GetGenePathwayRun then used syf:Synpcc7942 0655 to generate the pathway path:syf00195.

It is important to highlight that scientists can annotate the elements of a work ow-centric research object (along with the research object itself). They can specify the title of a research object, its purpose, its version, ownership, citations, etc. A more accurate form of annotation can be used to describe the elements of a research object by linking them to concepts from domain ontologies. In particular, this kind of annotation can be used to e ectively browse and query work ow templates.

Finally, work ow templates can be annotated in an abstract work ow template, which is a graph of abstract processes that are connected by data links. The abstract processes and their input and output parameters are labeled with concepts from underlying domain ontologies, e.g., [ 21, 22 ], which specify the tasks performed by the steps and the semantic domains of their parameters, respectively. An abstract worklfow template awf, which is used to annotate a given work ow template wf, has the same data ow topology as wf. The abstract processes that compose awf annotate the processes in wf, and the parameter domains in awf specify the semantic domains of the process parameters in wf. As an example, Figure 3-a illustrates an abstract work ow template that semantically describes the work ow template depicted in Figure 3-b. ProteinAcc to Gene and Gene to Pathway are two concepts that specify the tasks of the processes GetKeggGeneId and GetKeggPathway, respectively, whereas ProteinAccession, GeneId and Pathway are concepts that specify the domain of the input and output parameters of such processes. 3.1

Grounding Work ow-centric Research Objects Using Semantic Technologies Work ow-centric research objects are encoded using RDF5, according to a set of ontologies that we have made available6.

Following myExperiment packs, research objects use the Object Exchange and Reuse (ORE) model7, to represent aggregation. ORE de nes standards for the description and exchange of aggregations of Web resources. Using ORE, a work ow-centric research object is de ned as a resource that aggregates other resources, i.e., work ow(s), provenance, other objects and annotations. For example, the RDF turtle snippet illustrated below speci es that a research object identi ed by :wro aggregates a work ow template :pathway wf sp, a work ow run :pathway wf run, and an annotation :wf annot.

Example of a research object de ned as an ORE aggregation : wro a : WorkflowResearchObject , ore : Aggregation ; ore : aggregates : pathway wf sp , : pathway wf run , : w f a n n o t . : p a t h w a y w f s p a : WorkflowTemplate . : p a t h w a y w f r u n a : WorkflowRun . : w f a n n o t a ao : Annotation .

We also use the Annotation Ontology (AO)8, which provides a common model for annotating resources. This di ers from myExperiment packs, which use a vocabulary that is mapped to Open Annotation Collaboration (OAC)910. Several types of annotations are supported by the Annotation Ontology, e.g., comments, textual annotations (classic tags) and semantic annotations which relate elements of the research objects to concepts from underlying domain ontologies. As an example, the RDF turtle snippet below shows how the abstract work ow template illustrated in Figure 3-a can be speci ed using a named graph :pathway abs wf graph. It also shows how, using Annotation Ontology, such 5 http://www.w3.org/RDF 67 hhttttpp:/://w/wwww.wwf4e.voepr-epnroajerccth.oirvge/ws.iokir/gd/isoplraey//d1o.c0s//Rtoesce.ahrcthm+Olbject+Vocabulary+Specification 8 http://code.google.com/p/annotation-ontology 9 www.openannotation.org 10 Note that work is currently underway to align the two annotation vocabularies: http://www.w3.org/community/openannotation/ an abstract work ow template can be used to annotate the work ow template :pathway wf sp, which is depicted in Figure 3-b. Speci cally, a resource representing the annotation, :wf annot, is created to link the work ow template which is subject to annotation, :pathway wf sp, to the named graph specifying the corresponding abstract work ow template, :pathway abs wf graph. Example illustrating how a work ow template can be annotated using AO

The Lifecycle of a Work ow-Centric Research Object

We will now illustrate research object lifecycle through a small example that shows how all the resources contained in a research object are bundled as the scienti c experiment progresses. This example lifecycle is summarized graphically in Figure 4.

A research object normally starts its life as an empty Live Research Object, with a rst design of the experiments to be performed (which determines what work ows and resources will be added, by either retrieving them from an existing platform or creating them from scratch). Then the research object is lled incrementally by aggregating such work ows that are being created, reused or re-purposed, datasets, documents, etc. Any of these components can be changed at any point in time, removed, etc.

In our scenario, we observe several points in time when this Live Research Object gets copied and kept into a Research Object snapshot, which aims to re ect the status of the research object at a given point in time. Such a snapshot may be useful to release the current version of the research outcome of an experiment, submit it to be peer reviewed or to be published (with the appropriate access control mechanisms), share it with supervisors or collaborators, or for acknowledgement and citation purposes.

A snapshot may also contain a paper describing the research object in general and the experiment in particular, depending on the policies of the corresponding scienti c communication channel, e.g., workshop, conference or journal. Such snapshots have their own identi ers, and may even be preserved, since it may be useful to be able to track the evolution of the research object over time, so as to allow, for example, retrieval of a previous state of the research object, reporting to funding agencies the evolution of the research conducted, etc.

At some point in time, the research object may get published and archived, in what we know as an Archived Research Object, with a permanent identi er. Such a version of our research object may be the result of copying completely our Live Research Object, or it may be the result of some ltering or curation process where only some parts of the information available in the aggregation are actually published for others to reuse. As illustrated in Figure 4, a user can use an existing Archived Research Object as a starting point to his or her research, e.g., to repurpose it or its parts, in which case a new Live Research Object is created based on the existing Archived Research Object.

This is only one of the many potential scenarios that could be foreseen for the lifecycle of a work ow-centric research object and we are currently de ning di erent storyboards for their evolution. One important aspect to highlight is the fact that during its whole lifecycle, the research object is aggregating new objects. The annotation process during the lifecycle of experimentation allows the generation of su cient metadata about the research objects to support preservation and sharing. Therefore, when a scientists decides to preserve it most of the annotations that will be needed for that preservation process will be already available inside the research object. 5

Conclusions and Further Work

Scienti c work ows are used by scientists not only as computational units that encode scienti c methods that can be shared among scientists, but also to specify their experiments. In this paper we presented a research object model to capture all the needed information and data including the methods (work ows) and other elements: namely annotations, datasets, provenance of the work ow results, etc.

We showed how this model has been implemented using semantic technologies reusing existing vocabularies, so that scientists are now able to query and publish their experiments according to existing standards. As a result, experiments may be more interoperable, since they are recorded with the same general model to describe them; they can be reused more easily; and decay can be better handled by representing the information of the templates and the traces in an environment/execution independent manner.

The work reported in this paper is preliminary. Our ongoing work includes the design of an architecture for the management of work ow-centric research objects, based on the model presented in this paper, which is being implemented and made available in the Wf4Ever sandbox (http://sandbox.wf4everproject.eu/). We are also currently validating the model presented in this paper by creating research objects for existing work ows that are stored within the myExperiment repository. In doing so, we are examining issues that have to do with the decay of work ow, mechanisms for querying research objects, and scalability. As well as the technical challenges, we are aware that there are social challenges that need to be overcome to encourage scientists to adopt research object as a unit for publication, discovery and reuse of scienti c communications. In this respect, we started collaborating with scientists from the European projects BioVeL (Biodiversity Virtual e-Laboratory)11 and SCAPE (SCAlable Preservation Environments12).

Acknowledgements The research reported in this paper is supported by the Wf4Ever project (http://www.wf4ever-project.org), Project 270129 funded under EU FP7 Digital Libraries and Digital Preservation (ICT-2009.4.1). 11 http://www.biovel.eu 12 http://www.scape-project.eu

References

1. S. Bechhofer , I. Buchan , D. D. Roure , P. Missier , J. Ainsworth , J. Bhagat , P. Couch , D. Cruickshank , and Et Al. Why linked data is not enough for scientists . Future Generation Computer Systems , 2011 .

2. David Churches and Et Al. Programming scienti c and distributed work ow with triana services . Concurrency and Computation: Practice and Experience , 18 ( 10 ): 1021 { 1037 , 2006 .

3. Ewa Deelman and Et Al. Work ows and e-science: An overview of work ow system features and capabilities . FGCS , 25 ( 5 ): 528 { 540 , 2009 .

4. Lei Dou and Et Al. Scienti c work ow design 2.0: Demonstrating streaming data collections in kepler . In ICDE , pages 1296 { 1299 . IEEE Computer Society, 2011 .

5. Daniel Garijo and Yolanda Gil . A new approach for publishing work ows: Abstractions, standards, and linked data . In Proceedings of the Sixth Workshop on Work ows in Support of Large-Scale Science (WORKS'11) , held in conjunction with SC 2011 , Seattle, Washington, 2011 .

6. Yolanda Gil and Et Al. Mind your metadata: Exploiting semantics for con guration, adaptation, and provenance in scienti c work ows . In International Semantic Web Conference (2) , pages 65 { 80 . Springer, 2011 .

7. Yolanda Gil , Varun Ratnakar, Jihie Kim, Pedro Antonio Gonzalez-Calero, Paul Groth, Joshua Moody, and Ewa Deelman. Wings: Intelligent work ow-based design of computational experiments . IEEE Intelligent Systems , 26 ( 1 ), 2011 .

8. Darrel C. Ince , Leslie Hatton, and John Graham-Cumming. The case for open computer programs . Nature , 482 ( 7386 ): 485 { 488 , 02 2012 .

9. Bertram Ludascher, Ilkay Altintas, and Amarnath Gupta . Compiling abstract scienti c work ows into web service work ows . In SSDBM , pages 251 { 254 . IEEE Computer Society, 2003 .

10. Phillip Mates , Emanuele Santos, Juliana Freire, and Claudio T. Silva . Crowdlabs: Social analysis and visualization for the sciences . In SSDBM , pages 555 { 564 . Springer, 2011 .

11. Jill P. Mesirov . Accessible reproducible research. Science, 327 ( 5964 ): 415 { 416 , 2010 .

12. Paolo Missier , Satya S Sahoo, Jun Zhao, Carole Goble , and Amit Sheth . Janus: from work ows to semantic provenance and linked open data . Life Sciences , 6378 (i): 129 { 141 , 2010 .

13. Luc Moreau and Et Al. The open provenance model core speci cation (v1.1). Future Generation Comp . Syst., 27 ( 6 ): 743 { 756 , 2011 .

14. David Newman . The Building and Application of a Semantic Platform for an e-Research Society . PhD thesis , UNIVERSITY OF SOUTHAMPTON, 2011 . Submitted on October 2011 .

15. David Newman, Sean bechhofer , and David De Roure. myexperiment: An ontology for e-research . In Workshop on Semantic Web Applications in Scienti c Discourse in conjunction with the International Semantic Web Conference , 2009 .

16. Thomas M. Oinn and Et Al. Taverna: lessons in creating a work ow environment for the life sciences . Concurrency and Computation: Practice and Experience , 18 ( 10 ): 1067 { 1100 , 2006 .

17. Kevin R. Page , David De Roure, and Et Al. Rest and linked data: a match made for domain driven development ? In 2nd International Workshop on RESTful Design ( WS-REST 2011) held in conjunction with WWW 2011 , 2011 .

18. David De Roure and Et Al. The evolution of myexperiment . In e-Science 2010 . IEEE, 2010 .

19. David De Roure, Sean Bechhofer, Carole A. Goble , and David R. Newman . Scienti c social objects: The social objects and multidimensional network of the myexperiment website . In SocialCom/PASSAT. IEEE, 2011 .

20. David De Roure, Khalid Belhajjame, and Et Al. Towards the preservation of scienti c work ows . In Procs. of the 8th International Conference on Preservation of Digital Objects (iPRES 2011 ). ACM, 2011 .

21. Vuong Xuan Tran and Hidekazu Tsuji . Owl-t: A task ontology language for automatic service composition . In ICWS , pages 1164 { 1167 . IEEE Computer Society, 2007 .

22. Chris Wroe , Robert Stevens, Carole A. Goble , Angus Roberts, and R. Mark Greenwood . A suite of daml+oil ontologies to describe bioinformatics web services and data . Int. J. Cooperative Inf. Syst. , 12 ( 2 ): 197 { 224 , 2003 .