Semantic Integration for Model-based Life Science Applications

Tomasz Gubała

1,4

, Katarzyna Prymula

, Piotr Nowakowski

and Marian Bubak

2,4

Distributed Computing Environments Team, Academic Computer Centre Cyfronet AGH,

Nawojki 11, 30-950 Krakow, Poland

AGH University of Science and Technology, Department of Computer Science, Mickiewicza 30, 30-059 Krakow, Poland

Department of Bioinformatics and Telemedicine, Medical College, Jagiellonian University,

Sw. Łazarza 16, 31-530 Krakow, Poland

Informatics Institute, University of Amsterdam, Science Park 904, 1098 XH Amsterdam, The Netherlands

Keywords:

Semantic Modeling, Database Semantics, Genetics Models, In-silico Experimentation.

Abstract:

When delivering tools and solutions for modern e-Science applications, the proper transfer of domain knowl-

edge into information models is a crucial design step. In this work we describe the Semantic Integration

approach to modeling and transcribing complex science domain knowledge into well-structured information

models based on semantics. Apart from a conceptual study and presentation of related work, we also describe

how we applied the methodology to deliver a real-life solution for simulation of locations of active binding

sites in proteins – an important problem in bioinformatics. Our goal is to show how the Semantic Integration

technology can help deliver ready-to-use solutions for scientists.

1 INTRODUCTION

During the last decade we have witnessed the advent

of a new approach to scientiﬁc investigations using

computational facilities, based not only on large par-

allel supercomputers, but also employing application

servers, distributed systems and local workstations.

Researchers in areas such as modeling and simula-

tion (Guru et al., 2009) or computer-aided health-

care (Sloot et al., 2006) have been adopting a new

paradigm involving multiple steps, often performed

ad-hoc on local resources, with scripts, remote web

tools and database access helpers. The reason behind

this shift is twofold. First, it is about automation –

numerous preprocessing and output parsing and anal-

ysis tasks are becoming more and more automated to

streamline research work and increase effectiveness,

eliminating unnecessary effort and reducing the po-

tential for human error. Secondly, the ongoing devel-

opment of more agile and type-as-you-go program-

ming tools (including advanced integration mecha-

nisms for scientiﬁc workﬂows) has made it easier

for cross-domain educated scientists to create custom

tools for their daily research work. Automation and

dynamic experimentation also support another novel

scientiﬁc short-circuit technique – broad exploration

of solution spaces, where uninteresting avenues of re-

search can be automatically ﬁltered out.

This gradual rise in the effectiveness of scientiﬁc

experiments performed using computational software

and hardware comes at the expense of having to main-

tain a growing body of hand-tailored scripts, parsers

and analyzers, many of which are intended as one-off

helpers for a single purpose but later tend to become

(through constant ﬁxing and adjusting) very popular,

though convoluted, tools for a certain domain. Our

discussions with chemists and biologists indicate that

they spend a lot of their time digging through fora and

webpages for useful scraps of Python or Perl scripts.

While the very existence of such scripts (sometimes in

abundance) is a success in itself, further steps need to

be taken in terms of integration, standardizations and

interoperability. Research and deployment programs

like ESFRI

and EGI

(Bubak et al., 2012) are meant

to pave the way to integrated environments for future

system-level science.

A crucial part of our research on the development

of methods and tools for system-level science focuses

on the problem of integration and maintenance of

(sometimes quickly evolving) computational science

tools so that they may act as useful, natural toolkits for

European Strategy Forum on Research Infrastructures,

http://ec.europa.eu/research/infrastructure/

European Grid Infrastructure, http://www.egi.eu/

PL-Grid Project, http://www.plgrid.pl/

Gubała T., Prymula K., Nowakowski P. and Bubak M..

Semantic Integration for Model-based Life Science Applications.

DOI: 10.5220/0004426400740081

In Proceedings of the 3rd International Conference on Simulation and Modeling Methodologies, Technologies and Applications (SIMULTECH-2013),

pages 74-81

ISBN: 978-989-8565-69-3

 2013 SCITEPRESS (Science and Technology Publications, Lda.)

scientists. We closely observe the difﬁculties faced by

researchers who try to conduct complex e-science ex-

periments. From these observations we deﬁne the fol-

lowing challenge: what methodology a scientiﬁc pro-

grammer of in silico experiments should employ to in-

tegrate old and implement new software according to

the user expectations while avoiding one-off software

cludges (and thus worsening the situation for future

researchers)?

In the following sections we present an idea of se-

mantic integration (Gubala et al., 2009) which allows

scientiﬁc programmers and computational scientists

to integrate their workbench tools based on objects

and concepts from their own domain. The ﬂexibil-

ity of the method allows for constant improvements

while still maintaining a well-deﬁned semantic data

model that guarantees future maintainability and ex-

tensibility. Through an extensive case study of a spe-

ciﬁc application from the protein analysis domain we

prove that our proposal represents a viable solution to

the stated problem.

In order to position our work vis a vis current re-

search on web ontologies, shared taxonomies and on-

tology languages, we would like to elaborate on our

understanding of the crucial concept of ”semantics”.

In this paper, we deﬁne semantics as an assignment

of meaning to terms in the scope of a roughly de-

ﬁned domain. In other words, we consider seman-

tics to represent an answer to question: ”What does

this term mean to you?”, asked of a domain expert

(in line with Petrie (Petrie, 2009)). We do not equate

the semantic model with an ontological description,

nor do we use ontological languages directly in our

work. By doing that, however, we do not wish to

imply that ontological modeling is in any way infe-

rior to the taxonomy-based approach we use. We also

do not claim that ontologies are not applicable in our

work – merely that their application (or lack thereof)

is not crucial from the point of view of the seman-

tic integration approach proposed. While we prefer

to use non-ontological taxonomy modeling in our ap-

plication, we remain fully aware that using ontologies

(transcribed e.g. in OWL-DL) is also possible.

2 RELATED WORK ON

SEMANTICS USED

FOR INTEGRATION

Using domain semantics for integration of subsys-

tems is an approach often applied in (scientiﬁc) work-

ﬂow research. Since building workﬂows to model

data or control the ﬂow of an application is clearly

a method of integration, it can be based on domain

knowledge in the form of ontological annotations

(Corcho et al., 2006) or present in the workﬂow de-

scription itself (Bubak and Unger, 2007) (Kryza et al.,

2007). We argue, however, that using complex work-

ﬂow management and, intrinsically complex, onto-

logical modeling, might not be an option for smaller,

ad-hoc e-Science tools targeted by our research.

Substantial attention was also devoted to the in-

tegration of application assets, namely data sources

and computation elements. The variability and het-

erogeneity of data sources (in the form of databases or

medical instruments) may be tackled by applying se-

mantics as a single, common, sound data model (Can-

nataro et al., 2007). Apart from semantic annotation

of data sources, we may also perform semantic-based

dynamic query augmentation on such data, to deliver

an integrated database-like store (Kondylakis et al.,

2007). An example of semantic description of compu-

tation elements which act on data, usually to analyze

and transform it for scientiﬁc purposes, is Semantic

MOBY (P.M. Gordon, 2007). It attaches semantic in-

formation to a service so that users are able to quickly

ﬁnd and use the provided tools.

Both semantic integration of data (which models

the static part of an application) and semantic annota-

tion of services (to represent the dynamic aspect of an

application as a process) could be treated as parts of

models within the solution we propose (either the do-

main or the integration model). It is therefore possible

to combine such techniques with the idea of concept

and facet separation, presented above.

3 SEMANTIC INTEGRATION

IDEA

The core idea of the proposed methodology is to es-

tablish a ﬁrm basis of the integration process in a se-

mantic, domain-speciﬁc data model and to evolution-

arily extend it with an orthogonal integration model

as new tools and processes are integrated. This de-

composition into two different submodels, being the

core element of the proposed method, helps achieve

separation of concerns for each new application (re-

quirements of the domain against requirements of the

platform). If the domain expert (i.e. the researcher)

and the developer are different persons, the method-

ology assumes relatively tight collaboration between

the two – just like in any successful computational sci-

ence project. Starting with a set of elements of the sci-

entiﬁc experiment process, such as scripts, HPC ap-

plications and remote web applications, the program-

mer ﬁrst establishes the domain model.

SemanticIntegrationforModel-basedLifeScienceApplications

3.1 Modeling of Application Domain

The need for a domain model as a separate, indepen-

dent element, might not be so obvious at ﬁrst. When

approaching a typical problem of implementing and

conducting an in-silico experiment, one usually deals

with three to four tools which should be integrated

(perhaps as a workﬂow). It might be enough to estab-

lish a communication protocol based on implicit data

structures between these tools to make them work to-

gether; however there are two important factors to

consider. First of all, the developer may quickly dis-

cover additional elements which have to be integrated,

e.g. a persistence DB, a new web interface with its

corresponding server and client sides, perhaps a third-

party support library etc. Hence, as a result, the ﬁ-

nal application may have to incorporate not 3-4 but 4-

7 elements with nontrivial connections (for instance,

several domain tools may need to present something

using the web interface or may require some shared

persistence layer). Thus, the dedicated protocol ap-

proach becomes more challenging and results in less

code reuse.

Even more important is the fact that sooner or later

the tool may need to be extended, which is especially

true in modern life sciences. Therefore, expressing

the model implicitly in the protocols may severely im-

pact the maintainability and extensibility of the result-

ing solution.

For our domain model we propose a well-known

object-oriented approach. While some authors apply

ontologies to semantic domain modeling (Rodriguez

and Egenhofer, 2003), we have learned from our pre-

vious studies (Gubala and Hoheisel, 2006) that such

an approach is better suited for applications which

are very complex but also relatively stable (e.g. cri-

sis management systems for ﬂood control or pollution

monitoring). In situations where one deployment of a

tool serves for, at most, several months (sometimes

weeks) and then needs to be revised, ontology-based

taxonomies are too complex and not ﬂexible enough

to be an effective modeling solution. While ontology-

based description logic has its advantages over an ob-

ject model, we do not consider these as necessary to

properly model domain semantics. Another important

factor taken into consideration was the fact that pro-

grammers are far more familiar with object modeling

techniques (as they are part of any typical computer

science bachelor’s curriculum) compared to ontolog-

ical modeling – this relatively lenient learning curve

translates into better adaptability for recently formed

e-science teams.

Figure 1 shows a small part of our domain model

constructed for the HIV drug resistance domain in the

Figure 1: Example of a small part of a domain model (in

this case: HIV drug resistance) shown in a semantic inte-

gration notation. Specialization (hollow arrow) and com-

position (hollow rhombus) relations are applied, which is

typical of object-oriented modeling.

ViroLab project (Sloot et al., 2006). It deals with mu-

tations of HIV in human hosts and how these muta-

tions impact patient treatment options. The two most

critical relationships, inclusion and specialization, are

supported in our solution, in addition to primitive

datatypes and object arrays. The ﬁeld list has been

abridged for reasons of clarity.

3.2 Modeling of Platform Capabilities

The true novelty of our approach is the concept of an

integration model. In contrast to the domain model

(which represents concepts from a speciﬁc area of hu-

man knowledge), the integration model represents ev-

erything that is connected with the process of build-

ing and maintaining an application for this domain:

it should model all elements of the application as re-

quired by the persistence layer, communication proto-

cols, computing data structures and all aspects which

are orthogonal to, and not directly dependent on the

problem domain. In order to model integration as-

pects we apply a similar object-oriented approach, al-

though integration models are usually more dynamic

(i.e. more code is used to implement operations com-

pared to static, descriptive domain models). While el-

ements of the domain model are called ”concepts”, we

refer to elements of the integration model as ”facets”.

Figure 2 presents the integration model for the

example mentioned above. In the ViroLab Virtual

Laboratory there is need to store pieces of valuable

data computed in the course of experimentation and

to record the ”provenance” of such data by using a

specialized tool (Bali

s et al., 2008). Note both fea-

tures (recording results and dispatching provenance

events) are unrelated to the application domain (HIV

drug-impeding mutations) and instead stem from sys-

SIMULTECH2013-3rdInternationalConferenceonSimulationandModelingMethodologies,Technologiesand

Applications

Figure 2: HIV drug resistance example (continued): a por-

tion of the integration model for this application. The ”In-

sertion” concept from the domain model is automatically

outﬁtted with the capability of being both an ”Experimen-

tResult” and a ”ProvenanceEvent” – the programmer may

use either one, according to the context of the given ”Inser-

tion” instance.

tem requirements associated with the virtual labora-

tory software.

There are two reasons behind constructing the in-

tegration model as an independent entity. First, the

separation serves the purpose of clarity, and enables

parallel evolution of both models, as explained in the

next paragraph. Maintenance also becomes easier (es-

pecially when the models become complex) and can

even be entrusted to separate developers, as the rea-

sons for extending either model are connected with

different sets of requirements.

The other reason is to stress two different evolu-

tion paths of these two fundamental elements. The

domain model grows together with the scope of the

knowledge body being incorporated in the system,

which is mainly caused by applying existing use cases

or features to new aspects of the scientists’ research

work. On the other hand, the integration model grows

along with new capabilities (new use cases or require-

ments) being implemented. Since each new feature

may be (and usually is) initially implemented for the

most stable parts of the domain model, the exten-

sion process for both models remains, to some de-

gree, independent. Of course, both paths can never be

separated completely, yet we argue that the proposed

method of separation delivers the much sought-after

ﬂexibility of the constructed application.

3.3 ”Cartesian” Product of Facets

and Concepts for Integration

The ﬁnal prerequisite of practical application of our

solution is a suitable mechanism for joining both

Figure 3: A single ”Insertion” instance might be used in dif-

ferent contexts during the lifespan of the application. Here,

two different elements of the environment use publishing

and querying ”ExperimentResults” for integration based on

the common understanding of the meaning of ”Insertion”.

The second element may subsequently publish such results

in another context (as ”ProvenanceEvents”).

models. The semantic integration framework per-

forms the juxtaposition dynamically and automati-

cally, with no need for programmer intervention. As

presented in Figure 2, as soon as both models are de-

ployed (presumably as parts of a larger application),

the developer links it with a provided library (where

the implementation of the semantic integration for a

given language resides) which mixes them both, pro-

ducing what may be referred to as a ”Cartesian prod-

uct model”. Every ”concept” of the domain model

may assume any ”facet” of the integration model –

both models are meant to be orthogonal.

The usual programming experience with models

involves handling objects in memory. Here, the pro-

grammer uses the domain model to represent aspects

of knowledge in terms of objects, creating hierarchies,

declaring ﬁelds etc. As soon as the concepts have to

act in a speciﬁc area of the application, the developer

uses one of their facets to perform the requested op-

erations. Examples include storing data in a database,

sending messages to another part of the application

(or to an external entity), presenting the user with a

dedicated UI and similar. As shown in Figure 3, the

concept object instantly acquires all the ﬁelds and op-

erations of a speciﬁc facet, so that it can be stored,

sent or presented accordingly, along with all facet-

related information. We call this action ”publishing

with a facet”.

If the nature of a facet allows for subsequent re-

trieval of the underlying concept, it can be located via

its associated facet data (see Figure 3). The frame-

work performs a decoupling action so that the user

receives the concept object and the facet object sepa-

SemanticIntegrationforModel-basedLifeScienceApplications

rately. Now (for instance) the user may apply the con-

cept object to perform publication with a different as-

pect of the integration model. This mechanism proves

very valuable for middleware elements of applications

which tend to work in multiple aspects, such as per-

sisting data in a DB or ﬁle, relaying it as events, pub-

lishing with a remote API or presenting to the user.

Technically, the generation of the dynamic Carte-

sian product and decoupling are effected using the

dynamic self-inspection capabilities of the Ruby pro-

gramming language

. Similar techniques might be

applied in any other modern programming platform

providing introspection. In fact, scripting languages

such as Python or Ruby are frequent programming

tools of choice in modern life sciences applications.

4 CASE STUDY – THE

POCKETFINDER

APPLICATION

This study focuses on applying the proposed seman-

tic integration methodology and tools to the domain

of bioinformatic protein research. This section ex-

plains the idea behind the PocketFinder application

and provides a detailed view on how the mechanisms

described in the previous sections were used to deliver

a ﬁnal tool. In short, the main objective was to de-

velop a dedicated, efﬁcient data store, populate it with

a large amount of computation results and present the

outcome in an interactive way to scientists. The re-

sulting application is called PocketFinder and is in-

tended for internal use by the bioinformatics group

at the Jagiellonian University Collegium Medicum in

Krakow.

4.1 Identiﬁcation of Functional Sites

in Proteins

Understanding of how biological systems function is

the salient motivation for research in the ﬁeld of bio-

chemistry and molecular biology. Proteins are the key

performers responsible for function at the molecular

level. Revealing their functional sites, i.e. parts of

their structure directly involved in performing their

function, is an important objective. The term “func-

tional site” is commonly used in reference to binding

sites or catalytic residues. Accordingly, experimental

and computational methods for identifying and char-

acterizing functional sites of proteins are being inten-

we used Ruby for our implementation of the semantic

integration idea

0 20 40 60 80 100 120 140 160 180

Residue number

-0.008

-0.006

-0.004

-0.002

0.002

0.004

0.006

0.008

0.01

∆H

-0.008

-0.006

-0.004

-0.002

0.002

0.004

0.006

0.008

0.01

Figure 4: DNA-3-methyladenine glycosylase I. (top) Rib-

bon structure with semitransparent spheres representing

atoms colored by ∆

H values according to the scale pre-

sented in the proﬁle below. (Bottom) Hydrophobicity dif-

ference (∆

H) proﬁle.

sively developed (Vajda and Guarnieri, 2006; Weigelt

et al., 2008).

Here we present a procedure developed to opti-

mize FOD (Fuzzy Oil Drop model) – an in silico

method applicable to prediction of functional sites in

proteins (Brylinski et al., 2007). The prediction relies

on hydrophobicity difference values (∆

H) calculated

for each residue in a structure, based on FOD model

assumptions. The results for a given protein structure

are in the form of ∆

H proﬁle (Figure 4-top). The dis-

tribution of ∆

H values within a structure can be rep-

resented as well (Figure 4-bottom).

The main goal was to establish optimal parame-

ters of the FOD method for prediction of catalytic

residues. The calculations were performed on a test

set comprising 521 enzyme chains. The obtained ∆

values were treated as classiﬁers, discriminating cat-

alytic and non-catalytic residues. The actual class was

taken from CSA (Porter et al., 2004). As a result,

ROC curves were produced for each tested parameter

(in this case – hydrophobicity; Figure 5). Addition-

ally, areas bounded by the ROC curve (AUC) were

calculated.

4.2 Modeling Proteins and Proﬁles

We have applied the presented semantic integration

SIMULTECH2013-3rdInternationalConferenceonSimulationandModelingMethodologies,Technologiesand

Applications

0.2

0.4

0.6

0.8

0 0.2 0.4 0.6 0.8 1

TPR

FPR

Aboderin

Brylinski

Figure 5: ROC curves created by thresholding the results

obtained for 24 hydrophobicity scales. Two curves refer-

ring to the best and original hydrophobicity scales are high-

lighted.

approach to developing domain and integration mod-

els for the storage of protein hydrophobicity proﬁles.

Out of necessity, modeling such entities proceeded

according to an iterative, sequential process, as user

requirements were analyzed and reﬂected in the data

model. It should be noted that the solution presented

in the above sections is readily applicable to such iter-

ative development, as the application developer may

freely alter the domain model without affecting the

underlying integration model. Moreover, any change

in the domain model may be easily propagated to the

user interface layer by modifying a dedicated REST-

ful user interface representing the application domain

(in this case – protein functional site study). Thus,

the overall framework may be adapted to the require-

ments of individual application domains while still re-

taining overall genericity (as reﬂected by the integra-

tion model).

Within the speciﬁc scope of the PocketFinder ap-

plication, each protein is decomposed into multiple

aminoacids, each of which may be associated with an

arbitrary number of measurements. Moreover, each

protein may correspond to a speciﬁc hydrophobicity

value and binding proﬁle calculated on the basis of

multiple model types. This requirement gives rise to

a simple tree-like structure, which is depicted in Fig-

ure 6.

For PocketFinder we needed to prepare a data

store and ensure an efﬁcient method of storing the

results of multiple computations running in parallel.

Two requirements needed to be fulﬁlled at this stage:

the repository had to ﬁt in the available storage (thus,

the model had to store data in an efﬁcient manner)

and the resulting tool had to ensure rapid access to

the stored data.

Figure 6: The FOD domain model (the most important part)

represents the central concept of protein proﬁles.

While forgoing traditional ACID guarantees, the

semantic integration tool nevertheless enables devel-

opers to implement OOP-like data models and store

them in an efﬁcient manner. The ﬁnal data repos-

itory for the Protein Pockets case study application

contains approximately 1.5 GB of data, yet still en-

ables rapid browsing with the use of the research and

validation tools presented below. Certain require-

ments introduced at the implementation stage were

translated into additional integration model features –

of note is the publish as uniqueitem operation which

implements the standard upsert behavior present in

many data storage systems. Thus, the semantic in-

tegration tool admits domain-mandated extensions to

its core functionality without the need to re-engineer

or refactor the system as a whole.

4.3 Creation of PocketFinder Research

and Validation Tool

Having developed and deployed the domain data

model and user interfaces, we proceeded with the

implementation of a research tool which would per-

form the scientiﬁc calculations involved in the Protein

Pockets case study. The application assumed the form

of a series of scripts, all of which cooperate to provide

the required functionality:

• A selection of Ruby modules to populate the ini-

tial data store with information on proteins and

their constituent aminoacids;

• A Python experiment used to generate and execute

a sequence of scripts in a parameter-study mode,

where each of the launched scripts operates sep-

arately while retrieving data from the semantics

store;

• Perl scripts which implement the speciﬁc func-

tionality of the PocketFinder application, i.e.

SemanticIntegrationforModel-basedLifeScienceApplications

calculate measurement values for individual

aminoacids according to predeﬁned models, and

commit the results to the data store.

The RESTful domain interface (which is

application-speciﬁc and should be considered part

of the application itself), makes it easy to interface

the semantic integration tool from any programming

language equipped with a HTTP communications

library (which includes a vast majority of modern

languages). If we count shell scripting, the Pock-

etFinder application itself comprises four different

programming languages, all of which communicate

using the semantic integration tools.

5 OTHER APPLICATION AREAS

The presented solution was applied to several other

use cases, one of which was the HIV mutation and

drug resistance problem (Gubala et al., 2009) – in

fact, the requirements of that project were the driv-

ing force behind the semantic integration idea pre-

sented in this paper. Apart from the PocketFinder

application, we have also applied reﬁned method-

ologies in the computational chemistry domain (as

a semantic metadata engine in the InSilicoLab por-

tal (Kocot et al., 2012)

) and crisis warning and

management systems (as a semantic integration layer

for metadata and registries of dike sensors moni-

tored by the UrbanFlood EWS system (Balis et al.,

2011)

). Currently, we continouosly reﬁne and apply

the methodology to further science-related domains:

multiphisics and multiscale simulations in the scope

of MAPPER project (Rycerz and Bubak, 2011)

and

complex applications related to the Virtual Physio-

logical Human domain in the scope of VPH-Share

project (Nowakowski et al., 2012)

The development of these systems highlighted

good reusability of integration models (see 3.2). Once

deﬁned, facets such as ”UniqueInput”, ”Versioned-

Document” and ”JSONObject” tend to be applicable

to a variety of different realizations – this is due to

the relatively similar needs of users from different do-

mains, especially regarding the rather low level of the

software stack (persistence, presentation, communi-

cation). Here, programmers have greater freedom to

InSilicoLab e-science application portal, http://

insilicolab.cyfronet.pl/

Common Information Space, http://urbanﬂood.

cyfronet.pl/

Multiscale Applications on European e-Infrastructures,

http://www.mapper-project.eu/

VPH-Share Sharing for Healthcare - A Research Envi-

ronment, http://www.vph-share.eu/

adopt preferred solutions, so we opted for reusabil-

ity wherever the application requirements did not pre-

clude it. We believe that this observation validates our

core decision to separate the domain and integration

models.

6 SUMMARY

Summarizing, the ability to implement novel,

mission-critical tools in three different domains (in-

cluding HPC-based computational science and real-

time monitoring) in a six-month time frame by a team

of three programmers (not including end users who

contributed steady feedback) can be taken as evidence

for the extensive potential of the proposed method-

ology and accompanying technologies. It should be

noted, however, that when it comes to programmer

productivity or the agility of the software implemen-

tation process, it is difﬁcult to provide any convinc-

ing numerical measurements or metrics – particularly

ones that would involve semantics and their impact

on the process. Results need to be judged on the ba-

sis of common sense and past experience – yet the

outcome of our to-date research and development ef-

forts is a clear sign that the semantic integration idea

covered in this article is a worthwhile proposition for

e-science researchers and scientiﬁc developers alike.

ACKNOWLEDGEMENTS

The research presented in this paper has been par-

tially supported by the European Union within the

EU Project MAPPER (grant no. RI-261507) and the

EU Project VPH-Share (grant no. 269978). It was

also supported by the Polish national PL-Grid Project

POIG.02.03.00-00-007/08-00 and KI IET AGH grant.

REFERENCES

Bali

s, B., Bubak, M., Pelczar, M., and Wach, J. (2008).

Provenance tracking and querying in the virolab vir-

tual laboratory. In Proceedings of the 8th IEEE In-

ternational Symposium on Cluster Computing and the

Grid 2008 (CCGRID), pages 675–680. IEEE Com-

puter Society.

Balis, B., Kasztelnik, M., Bubak, M., Bartynski, T., Gubala,

T., Nowakowski, P., and Broekhuijsen, J. (2011). The

urbanﬂood common information space for early warn-

ing systems. In Proceedings of the International Con-

ference on Computational Science ICCS 2011, vol-

ume 4, pages 96–105. Procedia Elsevier.

SIMULTECH2013-3rdInternationalConferenceonSimulationandModelingMethodologies,Technologiesand

Applications

Brylinski, M., Prymula, K., Jurkowski, W., Kochanczyk,

M., Stawowczyk, E., Konieczny, L., and Roterman,

I. (2007). Prediction of functional sites based on the

fuzzy oil drop model. PLoS Comput Biol, 3(5):e94.

Bubak, M., Szepieniec, T., and Wiatr, K., editors (2012).

Building a National Distributed e-Infrastructure -

PL-Grid - Scientiﬁc and Technical Achievements.

Springer.

Bubak, M. and Unger, S., editors (2007). KWf-Grid -

The Knowledge-based Workﬂow System for Grid Ap-

plications. Academic Computer Centre CYFRONET

AGH.

Cannataro, M., Guzzi, P., Mazza, T., Tradigo, G., and Vel-

tri, P. (2007). Using ontologies for preprocessing and

mining spectra data on the grid. Future Generation

Computer Systems, 23(1):55–60.

Corcho, O., Alper, P., Kotsiopoulos, I., Missier, P., Bech-

hofer, S., and Goble, C. (2006). An overview of s-

ogsa: A reference semantic grid architecture. Web Se-

mantics: Science, Services and Agents on the World

Wide Web, 4(2):102–115.

Gubala, T., Bubak, M., and Sloot, P. M. (2009). Semantic

Integration of Collaborative Research Environments,

chapter XXVI, pages 514–530. Information Science

Reference IGI Global.

Gubala, T. and Hoheisel, A. (2006). Highly dynamic work-

ﬂow orchestration for scientiﬁc applications. In Core-

GRID Intergation Workshop 2006 (CIW06), pages

309–320. ACC CYFRONET AGH.

Guru, S. M., Kearney, M., Fitch, P., and Peters, C. (2009).

Challenges in using scientiﬁc workﬂow tools in the

hydrology domain. In 18th World IMACS and MOD-

SIM09 Int. Congress on Modelling and Simulation,

pages 3514–3520. Modelling and Simulation Society

of Australia and New Zealand.

Kocot, J., Szepieniec, T., Harezlak, D., Noga, K., and

Sterzel, M. (2012). InSilicoLab Managing Com-

plexity of Chemistry Computations, pages 265–275.

Springer.

Kondylakis, H., Analyti, A., and Plexousakis, D. (2007).

Quete: Ontology-based query system for distributed

sources. Lecture Notes in Computer Science,

4690:359–375.

Kryza, B., Slota, R., Majewska, M., Pieczykolan, J., and Ki-

towski, J. (2007). Grid organizational memory – pro-

vision of a high-level grid abstraction layer supported

by ontology alignment. Future Generation Computer

Systems, 23(3):348–358.

Nowakowski, P., Bartyski, T., Gubaa, T., Harlak, D.,

Kasztelnik, M., Meizner, J., and Bubak, M. (2012).

Managing cloud resources for medical applications.

In Proceedings of Cracow Grid Workshop 2012

(CGW’12), pages 35–36. ACC Cyfronet AGH.

Petrie, C. (2009). The semantics of ”semantics”. IEEE

Internet Computing, 13(5):96–95.

P.M. Gordon, Q. Trinh, C. S. (2007). Semantic Web Ser-

vice provision: a realistic framework for Bioinformat-

ics programmers. Bioinformatics, 23(9):1178–1180.

Porter, C. T., Bartlett, G. J., and Thornton, J. M. (2004). The

Catalytic Site Atlas: a resource of catalytic sites and

residues identiﬁed in enzymes using structural data.

Nucleic Acids Res, 32(Database issue):D129–D133.

Rodriguez, M. and Egenhofer, M. (2003). Determiniing

semantic similarity among entity classes from differ-

ent ontologies. IEEE Transactions on Knowledge and

Data Engineering, 15(2):442–456.

Rycerz, K. and Bubak, M. (2011). Building and Run-

ning Collaborative Distributed Multiscale Applica-

tions, chapter 6, pages 111–130. J. Wiley and Sons.

Sloot, P. M. A., Tirado-Ramos, A., Altintas, I., Bubak, M.,

and Boucher, C. (2006). From molecule to man: De-

cision support in individualized e-health. Computer,

39(11):40–46.

Vajda, S. and Guarnieri, F. (2006). Characterization of

protein-ligand interaction sites using experimental and

computational methods. Curr Opin Drug Discov De-

vel, 9(3):354–362.

Weigelt, J., McBroom-Cerajewski, L. D. B., Schapira, M.,

Zhao, Y., and Arrowmsmith, C. H. (2008). Structural

genomics and drug discovery: all in the family. Curr

Opin Chem Biol, 12(1):32–39.

SemanticIntegrationforModel-basedLifeScienceApplications