=Paper=
{{Paper
|id=None
|storemode=property
|title=iPlant SSWAP (Simple Semantic Web Architecture and Protocol) Enables Semantic Pipelines for Biodiversity
|pdfUrl=https://ceur-ws.org/Vol-979/WS_s4biodiv2013_paper_9.pdf
|volume=Vol-979
}}
==iPlant SSWAP (Simple Semantic Web Architecture and Protocol) Enables Semantic Pipelines for Biodiversity==
https://ceur-ws.org/Vol-979/WS_s4biodiv2013_paper_9.pdf
iPlant SSWAP (Simple Semantic Web Architecture
and Protocol) enables semantic pipelines for
biodiversity
Damian D. G. Gessler1, Blazej Bulka2 , Evren Sirin2, Hans Vasquez-Gross3,
John Yu3, Jill Wegrzyn3
1
The iPlant Collaborative, University of Arizona, Tucson, AZ, U.S.A.
dgessler@iplantcollaborative.org
2
Clark and Parsia, Washington, D.C., U.S.A.
{blazej,evren}@clarkparsia.com
3
University of California, Davis, CA, U.S.A.
{havasquezgross,jjsyu,jlwegrzyn}@ucdavis.edu
Abstract. Real-time response is a basic characteristic of the Web. Yet
semantic reasoning at transaction-time supporting real-time response
remains challenging. Here we report how the iPlant Semantic Web
Platform uses SSWAP (Simple Semantic Web Architecture and Protocol;
http://sswap.info) for transaction-time reasoning, service discovery,
workflow construction, and execution. The platform enables users at web
sites, such as TreeGenes’ DiversiTree and CartograTree, to select data
and use it for real-time semantic discovery into a knowledge base of
semantic web services. The platform uses first-order, description logic
reasoning and just-in-time ontologies to allow users to drag-n-drop
independent, distributed semantic web services into a semantic pipeline.
This enables biodiversity research using data sets from TreeGenes,
FLUXNET (Ameriflux), WorldClim, and TRY-DB integrated under a
common web front-end called CartograTree. Scientific use cases are for
tree scientists to associate phenotype and/or environmental traits with
underlying genotypes in geo-referenced forest trees across a distribution
of Web resources.
Keywords: Semantic Web Services, SSWAP, iPlant, TreeGenes,
DiversiTree, CartograTree, OWL
1 Introduction
Bioinformatic software exhibits long-tail characteristics: a relatively small
number of programs and web sites are widely used (e.g., [1,2]), while a much
2
larger number are used by varied audiences for specialized applications (e.g.,
[3,4]). The iPlant Collaborative [5] seeks to enable data-driven scientific
integration, both within the enterprise and across Web resources, including
widely used programs of general interest and niche programs for specific needs.
Emphasis is on having software layers handle data and service syntax and
semantics (including independently developed and maintained long-tail
offerings) thereby freeing the scientist to focus on data and service discretionary
use. To achieve this, iPlant is using SSWAP (Simple Semantic Web
Architecture and Protocol [6]) in a drag-n-drop semantic pipeline motif with
third-party Web site integration. In this paper we report how a collaboration
with TreeGenes [7] enables biodiversity applications in forest genetics.
Exemplary applications in land management and biodiversity include the
identification of specific genotypes that may be best suited for reforestation, or
the development of strategies for tree migration as it relates to climate change.
In both cases, genotypes that influence traits such as cold-hardiness, drought-
tolerance, and disease resistance can be examined in relation to environmental
characteristics of target regions including elevation, soil composition, and
precipitation.
2 The Platform
Architecture The iPlant Semantic Web Platform is a Web architecture of
four distributed actors: i) providers of services; ii) consumers of services; iii)
ontology severs; and iv) a semantic Discovery Server (pipeline-maker and
match-maker). Data—be it unstructured, semi-structured, or structured (e.g., as
in relational database stores)—enters the system via a service interface layer;
i.e., the platform does not operate on raw data per se, but via service interfaces,
the invocations of which yield access to, and transformations of, data. This
service interface layer is key to enabling distributed data to be integrated
“rationally” under a first-order description logic protocol.
SSWAP (Simple Semantic Web Architecture and Protocol) SSWAP is a
100% W3C OWL DL-compliant light-weight protocol of five classes and 12
properties. It allows services to describe what they are, the types of data they
consume, and the types of data they produce. The protocol’s ontology in its
entirety is at [8]. The five classes correspond to: i) the service Provider, ii) the
service itself, called a Resource, iii) a data structure construct called a Graph, iv)
input data (a Subject), and v) output data (an Object). SSWAP is the service
analog of the fundamental RDF data model of mapping a subject to an object via
a property; in the cases of SSWAP, the protocol maps a Subject to an Object via
the implicit operation of a service (the Resource). Subject and Object instances
may be URIs, thereby allowing for indirection and non-serialization of data, or
3
they may identify data structures of arbitrary OWL sub-graphs, with properties
and serialized data. Instances of Resource, Subject, and Object may be annotated
with user-defined ontologies and thus are “unlimited” in domain scope; the
protocol simply defines the scaffold. Services may have multiple Subjects
mapping to multiple Objects. A protocol description of a service is called an
RDG (Resource Description Graph). An HTTP GET on the Resource URL of
the RDG returns the RDG in W3C-compliant OWL RDF/XML. Because service
descriptions are just text documents retrievable by a simple GET, they are
readily available for search engine traversal and viewable by browsers1. An
RDG with input data creates an RIG (Resource Invocation Graph). An HTTP
POST of the RIG or a GET with ontology term=value assignments in the query
string invokes the service. An RIG with output data is called an RRG (Resource
Response Graph). Thus SSWAP creates an ecosystem of protocol graphs, all
sharing a canonical model, with a common syntax (OWL RDF/XML), under a
common services’ semantic (SSWAP), amendable to customization by user
semantics (adding ontology terms to the Resource, Subject, and Object).
SSWAP is a wrapper technology, so it can semantically enable legacy and non-
semantic services. Notably, a SSWAP service description yields the service
amenable to automated semantic discovery, invocation, and response.
Semantic Querying A service’s protocol description encapsulates the
information needed for its discovery and invocation. Thus one can consider any
putative RDG as a query graph (called an RQG: Resource Query Graph) into a
knowledge base of all RDGs. For semantic querying, we find all services for
which the RQG’s: i) Resource is a subclass, and ii) Subject is a super-class, and
iii) Object is a subclass, of any service in the knowledge base. Subsumption
reasoning covers arbitrary complex, inferred, anonymous classes. The resultant
services, and only these services, are guaranteed to be of the type of service
queried (or more specialized), to operate on the input data (or generalizations of
it), and return data of the requested output type (or specializations of it). This
allows us to use a reasoner for match-making based on the output of one service
being logically sufficient for the input of another. Thus reasoning is used to
examine service descriptions, input data types, and output data types, to enable
semantic matching with published services.
Constructing semantic pipelines At http://sswap.info, a Web front-
end to a backend pipeline manager allows users to connect services into
pipelines. Pipelines are built on-demand by using transaction-time reasoning to
aid the user in building a workflow of distributed services.
Start with a lexical search Users at http://sswap.info may search for
services using keywords. Upon selecting a service and adding it to a new
1
Visit http://sswap.info, search for a service, click on ‘Service URI’ to view
the RDG, or visit http://sswap.info/api/makeRDG for examples.
4
pipeline via web-based drag-n-drop, we present the user with all downstream
services that can operate on the upstream service via semantic querying as
described above. In this manner, the user can build a pipeline of services. For
each service, we reason over the service’s RDG to determine its necessary and
sufficient conditions, and based on this construct on-demand a custom user
dialog that allows the user to enter the service’s required and optional
parameters, if any. In a similar manner the Subject is examined, and the user
may upload data to be ontologically tagged via the RDG. The protocol declares
a datatype property sswap:inputURI2 which allows service providers to write
custom Web pages to solicit user input for their services. If sswap:inputURI
resolves to a Web page, the platform will present that page to the user in
addition to allowing the user to use the auto-generated, custom user dialog.
Start with data launched from a web site We provide a Javascript snippet that
allows any webmaster to add a “sswap.info” button to their web pages. We call
this Web Discovery. We provide a service to allow the Web master to package
or reference the data using JSON (see /api)3. Upon the user pressing the Web
Discovery button, the JSON is sent to our Discovery Server, where we translate
it into an RDF/XML RQG, perform semantic querying, and present the user
with a new pipeline preloaded with their data and the semantic results of all
matching candidate downstream services.
Start with the results from previous pipelines Because the last service in a
pipeline returns a standard RRG, this can be used to start a new pipeline. In this
manner, a pipeline can seed new pipelines. Data is private, but pipelines may be
published for public use and are semantically discoverable like services. In this
manner, we grow a database of user-built combinations of Web distributed
services; this has deep social networking value. We note that public sharing of
pipelines does not imply unregulated execution of services: any service is free to
gate-keep resources with logins, HTTPS, and so forth.
Pipeline invocation is orchestrated, but execution is distributed RDGs
represent published SSWAP services that are offered by third-parties on the
Web. When the user initiates a pipeline, we coordinate the invocation and
callback of services, but do not ourselves execute the services: the services run
independently, asynchronously on their host machines. Downstream services
retrieve the upstream RRG from the upstream service with a token and convert it
to an RIG without passing through our servers. In this manner we are not privy
to non-serialized data being transferred between services, thereby maintaining
an important privacy safe-guard.
Transaction-time reasoning SSWAP graphs (RDGs, RIGs, RRGs, and
RQGs) are small documents of a few dozen lines of W3C OWL [DL]
2
sswap: prefix resolves to http://sswapmeet.sswap.info/
3
Relative URLs are RESTful endpoints on http://sswap.info/
5
RDF/XML that typically expand to a few thousand triples after first-order
reasoning. We use reasoning in four places: i) when Providers publish their
RDGs with us, we resolve ontology terms by dereferencing them on the Web;
we then infer over the closure RDG and store the resulting inferred graph in a
triple-store [9]. We use a combination of transaction-time reasoning at
publication time and offline processing to maintain the knowledge base; ii)
when users initiate Web Discovery from a web site by sending us a JSON
representation of an RRG, we resolve the RRG, convert it to a RQG, and
execute transaction-time semantic querying; iii) when users build pipelines we
reason during the transaction process to satisfy semantic querying and other
pipeline duties; iv) when third-party services receive an RIG they need to
process the request and return a RRG that complies with the logical contract of
their RDG. We provide a kit (/sdk) that allows third-parties to run their own
servlet reasoner to handle transaction-time reasoning to process requests.
Pipeline management Control is architected as three separate components: i)
we use Vaadin [10] to offer a RIA (Rich Internet Application) enabling an
intuitive, drag-n-drop user experience; ii) communication to the backend is
performed by a 100% RESTful JSON API, making heavy use of idempotent
HTTP GETs and PUTs. This means that a user may start building a pipeline,
bookmark it, close their browser, and open it anywhere, anytime, and continue
their work. It means that users may begin long-running pipelines, and return at
their convenience with a different browser and Web session; iii) the pipeline
manager communicates with the Discovery Server via a RESTful API.
Platform APIs We wrote ~185,000 lines of open-source Java code to build a
platform, Java API (/javadocs), and helper services. We use the Java API
internally, and package it as part of our SDK (Software Development Kit) so
anyone may write their own SSWAP services (/sdk). Many developers are
fluent in JSON, but not in OWL RDF/XML, so we wrote a RESTful translator
that allows SSWAP graphs and user ontologies to be written in JSON and then
translated to OWL RDF/XML (/api; see also /make and [11]). We expose
Discovery Server engagements as RESTful endpoints (/wiki/api).
3 Ontologies
A challenge for the semantic web services is how to enable and incorporate
distributed ontologies. We enable the use of user-defined OWL ontologies to
allow services to describe their data, and to allow clients to query and engage
said services.
Just-In-Time ontologies We used Smart GWT [12] to write an application
that allows anyone to host their ontologies on our servers [11]. Users register for
a free iPlant account and may create and administer new ontologies (called
6
“namespaces”). Users build ontologies term-by-term using a JSON syntax [13],
translate them to RDF/XML with the press of a button, and publish them on-
demand. Terms are separately dereferenceable and immediately available to
anyone on the web. Just-In-Time ontologies lower the barrier to entry for
creating and using small, agile ontologies, but they are not required: ontologies
residing anywhere on the web may be freely used, subject to byte and time
limits during transaction processing. Ontological statements (e.g., definitions
and relation to other terms) are read and used in reasoning if dereferencing term
URIs returns OWL RDF/XML statements.
Support for “large” legacy ontologies: module extraction with BioPortal
BioPortal [14] is a major repository funded by the National Center for
Biomedical Ontology. It contains over 320 ontologies, and over 180 OWL
ontologies. We use the method of [15,16] to process each OWL ontology offline
to generate “atoms,” such that at transaction-time we can compute the subset of
statements (called a “module”) that are necessary and sufficient for complete
entailment over any subset of terms (called a “signature”). Importantly, for
moderate sized signatures the module is often much smaller than the ontology
itself [15], thus lending it as a key approach to bringing large, legacy ontologies
to transaction-time applications in the semantic Web. Currently, ontology
modularization is available as a service at /modularize. As of this writing,
we are implementing a strategy to incorporate it into transaction-time processing
but this is not yet part of the larger platform.
Ontologies enable semantic querying and reasoner-assisted semantic
pipeline construction When we process a SSWAP graph, we extract ontology
terms and dereference them to retrieve their OWL statements. If these
documents themselves contain terms, we dereference those, and continue this
cascade until closure is achieved, subject to traversal depth, byte, and time
limits. For Web Discovery and pipeline construction we then use Semantic
Querying (described above) to find matches between data and/or the output
semantics of the upstream service and the input semantics of all putative
downstream services. Subsumption determination is performed at transaction
time, so axiomatic subsumption claims (e.g., rdfs:subClassOf) are
supported but not required: the reasoner uses transaction-time classification to
determine subsumption. Note that it is the SSWAP protocol that makes this
possible, because the protocol ensures that the subject and object semantics of
RDGs, RIGs, RRGs, and RQGs are comparable.
7
4 Integrating Enterprise, HPC, and the Semantic Web for
Biodiversity
Enterprise resources TreeGenes [7] is a large biological resource serving
over 2500 forest geneticists from over 800 organizations. It contains data from
15 yrs on over 1200 species, including genomic, phenotypic, and other data. We
wrote 11 SSWAP services to expose slices of this data and added SSWAP Web
Discovery to TreeGenes’ DiversiTree [17]. For geographically-oriented tree
scientists, we wrote a mapping tool called CartograTree [18,19]. Researchers
can search specific geographic regions, tree species, phenotypes, or
environmental parameters and customize their analysis accordingly. We enabled
CartograTree with SSWAP Web Discovery so that scientists can launch directly
into semantic discovery. The iPlant Collaborative serves over 7500 scientists
with enterprise-class and High Performance Computing (HPC) resources,
petabyte-scale storage, and other resources. We wrote semantic pipeline support
to engage HPC XSEDE resources [20] and used SSWAP to semantically wrap
10 resources in the domain of multiple sequence alignment and phylogenetic
tree reconstruction.
Biodiversity The DiversiTree/CartograTree/SSWAP integration is driven by
questions arising from climate change, disease resistance, and conservation.
Knowledge of the adaptive genetic potential of forest tree populations is
critically important for evaluating their vulnerability to a changing climate [21].
Forests are key to sequestering carbon and consequently contribute an important
role to mitigating or reinforcing the impacts of climate change. Healthy forests
provide fundamental habitat for valued biodiversity and essential ecosystem
services in the form of global carbon cycling, clean water and air, fiber, and
recreation. Sustaining healthy forests in the face of climate change is a central
challenge for resource management [22]. Towards this goal, researchers are
examining candidate loci to understand how individuals and populations are
impacted by environmental factors. Specifically, a fusion of population genetics
and landscape ecology to layered geographic information systems allows for
focused studies of how landscape features affect genetic variation [23-25].
Experimental design often focuses on first identifying candidate genes under
selection from geoclimatic factors, determining their allelic diversity, and testing
for associations between trees’ genotype, phenotype, and the environment.
CartograTree connects the TreeGenes’ repository of genotype and sequence data
to environmental and phenotypic resources. TreeGenes houses approximately
901,000 sequences, 24 million genotypes, and 20,000 phenotypes on individuals
from over 1,200 different forest tree species. Sequencing includes either Sanger-
based or next-generation approaches, and used to identify polymorphisms in
small populations. The polymorphisms are then validated in larger populations
8
through the use of high-throughput genotyping assays. In many cases, genotyped
trees are phenotyped for various traits. Barcode identifiers assigned during
sample collection are maintained through DNA extraction, sequencing,
genotyping, and phenotyping, while also associating trees with their geo-
referenced coordinates. The external sources supplying environmental and
phenotypic data include relevant portions of the FLUXNET (Ameriflux) [26],
WorldClim [27], and TRY-DB [28] repositories. Ameriflux represents 81
remote sensing sites across North and South America; WorldClim is a
compilation of five different climate databases covering the globe; TRY-DB
enhances phenotypic data with approximately 80,366 geo-referenced phenotypic
records represented by 368 species. Within CartograTree, specific queries and
filters are available to select by genus, species, or phenotype of interest. The
phenotypic selections include economically relevant traits, disease evaluations,
and developmental metrics. The map portion of the interface gives users the
option to select regions of interest, and capture the associated environmental
data, such as slope, elevation, precipitation, seasonal temperatures, and more.
From this, scientists can send selected data for SSWAP Web Discovery, for
example, to perform multiple sequence alignment and phylogenetic tree
reconstruction on high performance computing clusters. A full description of
CartograTree and SSWAP is published at [19]. Association studies are
facilitated through the ability to create flat files based on the common
phenotypic or environmental evaluations for a selection of trees. The results of
these studies are aimed at improving land-management decisions through the
identification of genotypes that will thrive in specific environments; information
that is necessary for reforestation, disease resistance, and climate change.
5 Conclusion
Semantics and biodiversity is still in its nascent years. Our work is focused on
a division of scientific labor between domain-specific information resources
such as TreeGenes, infrastructural resources such as iPlant, high performance
computing assets such as underlying the phylogenetic applications available on
XSEDE, and the larger Web. iPlant’s Semantic Web Platform is developed as
the technological conduit for integration across these resources. It uses
transaction-time first-order description logic reasoning to allow semantic web
services to be discovered, connected, and invoked via a simple drag-n-drop web
interface. TreeGenes, DiversiTree, and CartograTree offer an initial foray into
the use of these technologies for forest tree biodiversity.
Acknowledgements We thank Pavel Klinov for work on ontology
modularization and Yan Kang for work on the Just-In-Time Ontology editor.
9
This work was supported by NSF grants for the iPlant Collaborative (#DBI-
0735191) and SSWAP (#0943879).
6 References
1. http://www.ncbi.nlm.nih.gov/sites/gquery
2. Altschul S.F., Gish W., Miller W., Myers E.W. and Lipman D.J. Basic local
alignment search tool. J. Mol. Biol. 215: 403-410 (1990).
3. https://pods.iplantcollaborative.org/wiki/display/DEapps
/List+of+Applications
4. Goecks, J, Nekrutenko, A, Taylor, J and The Galaxy Team. Galaxy: a
comprehensive approach for supporting accessible, reproducible, and transparent
computational research in the life sciences. Genome Biol. Aug 25;11(8):R86 (2010).
5. Goff, S. A. et al., The iPlant Collaborative: Cyberinfrastructure for Plant Biology.
Frontiers in Plant Science 2, 1 - 16 (2011). DOI: 10.3389/fpls.2011.00034.
6. Gessler, D.D.G., Schiltz, G.S., May, G.D., Avraham, S., Town, C.D., Grant, D.,
Nelson, R.T.: SSWAP: A Simple Semantic Web Architecture and Protocol for
semantic web services. BMC Bioinformatics, 10:309, pp. 1-21 (2009).
7. http://dendrome.ucdavis.edu
8. http://sswapmeet.sswap.info/sswap; see also
http://sswapmeet.sswap.info/jit/sswap. Also included at the URL is
an additional class and property for asynchronous service invocation.
9. Pellet + Stardog; see http://stardog.com
10. http://vaadin.com
11. http://sswapmeet.sswap.info
12. http://smartclient.com
13. http://sswap.info/api/JSONSyntax
14. http://bioportal.bioontology.org
15. Del Vescovo, C., Gessler, D.D.G., Klinov, P., Parsia, B., Sattler, U., Schneider, T.,
and Winget, A.: Decomposition and Modular Structure of BioPortal Ontologies, In:
ISWC, LNCS, 7031: pp. 130-145 (2011)
16. Klinov, P., Del Vescovo, C., Schneider, T.: Incrementally updateable and persistent
decomposition of OWL ontologies. In: Proceedings of OWL: Experiences and
Directions Workshop 2012. Klinov, P., Horridge, M. (eds.) Heraklion, Crete,
Greece, May 27-28, 2012. CEUR Workshop Proceedings 849 CEUR-WS.org 2012.
17. http://dendrome.ucdavis.edu/DiversiTree
18. http://dendrome.ucdavis.edu/cartogratree
19. Vasquez-Gross, H.A., Yu, J.J., Figueroa B., Gessler D.D.G., Neale D.B., Wegrzyn
J.L.: CartograTree: connecting tree genomes, phenotypes and environment.
Molecular Ecology Resources, (2013) doi: 10.1111/1755-0998.12067
20. https://www.xsede.org
21. Neale D.B., Kremer A.: Forest tree genomics: growing resources and applications.
Nature Reviews Genetics 12, pp. 111–122 (2011)
10
22. Peterson, D.L.; Halofsky, J.E.; Johnson, M.C.: Managing and adapting to changing
fire regimes in a warmer climate. In: McKenzie, D.; Miller, C.; Falk, D., (eds.) The
landscape ecology of fire. New York: Springer: Chapter 10, pp. 249–267 (2011)
23. Manel, S., Joost, S., Epperson, B.K. et al.: Perspectives on the use of landscape
genetics to detect genetic adaptive variation in the field. Molecular Ecology, 19, pp.
3760–377 (2010)
24. Manel, S., Schwartz, M.K., Luikart, G., Taberlet, P.: Landscape genetics: combining
landscape ecology and population genetics. Trends Ecol. Evol. 18, 189-197 (2003)
25. Feder, M., Mitchell-Olds, T.: Evolutionary and ecological functional genomics.
Nature Genetics Reviews, 4, pp. 649-655 (2003)
26. Baldocchi, D., Falge, E., Gu, L.H., et al.: FLUXNET: a new tool to study the
temporal and spatial variability of ecosystem-scale carbon dioxide, water vapor, and
energy flux densities. Bulletin of the Amer. Meteor. Soc., 82, pp. 2415–2434 (2001)
27. Hijmans, R.J., Cameron, S.E., Parra, J.L., Jones, P.G., Jarvis, A.: Very high
resolution interpolated climate surfaces for global land areas. International Journal of
Climatology, 25, pp. 1965–1978 (2005)
28. Kattge, J., Diaz, S., Lavorel, S. et al.: TRY - a global database of plant traits. Global
Change Biology, 17, pp. 2905–2935 (2011)