BieColl - Bielefeld eCollections

Öztürk, Ahmet R.

  Proceedings of CELLmicrocosmos neXt 2014 (DOI: 10.2390/biecoll-next2014-5)  12
Web-based Visualization with D3
Ahmet R. Öztürk*
Bioinformatics Group, Genome and Stem Cell Center, Erciyes University, Kayseri, Turkey
1 Current Visualization Approaches and D3
Visual representation of complex data reveals the motifs and patterns which cannot be
easily understood by basic statistical approaches. However, in case of the analysis of real
data,  the  phenomenon of  interest  usually  involves  many  parameters  and a  researcher
needs many different plots to visualize the diverse aspects of analysis results. In addition,
current paradigm of visualization is based on displaying static images on a 2D surface. It
might be a computer screen, a wall, a piece of paper, or any kind of material that provides
a support for the image.
A solution for displaying complex data visualizations is to make visualization interactive
through a computer or a smart device. This gives the user much more flexibility for looking
at different aspects of the data with little effort. There are many standalone tools for
specific  visualization  purposes  such  as  Cytoscape  and  Gephi  for  network  visualization
whereas standalone software has some serious limitations for interaction flexibility and
sharing  the  visualizations.  Web  technologies  provide  a  great  deal  of  opportunity  for
interactive data visualization and among many JavaScript library alternatives, D3 is well
ahead [1]. D3 is a JavaScript library [2] that allows programmers to modify and transform
any part of a web page and associate HTML elements with data. This results in interactive
and dynamic graphs with a very wide range of layout options through a modern browser. 
1.1 Visual consistency of D3
Another advantage of using D3 is object constancy. Usually, when you generate a plot with
different elements, positions of each object are calculated from scratch using a built-in
algorithm. This  independent nature of  calculations  prevents  continuity of  the graph in
response  to  a  change  of  a  parameter  in  underlaying  data.  This  is  a  counterintuitive
approach regarding our cognitive expectations. The visual continuity is achieved through
animated transitions. There are examples showing how visual continuity helps the user to
easily follow changes in response to a parameter change [3].
A suitable example for displaying the results of calculations of graph element positions
without considering visual continuity is a BLOSUM network using specific BLOSUM matrices.
Consider the network graphs below in Figure 1, with different BLOSUM matrices of 62% and
45% similarities, respectively.
On the graphs shown in Figure 1, when the matrix parameters change, positions of each
node are calculated from the beginning. However, when you generate the initial network
with D3 and enable the user to select the new matrix to be displayed, user can easily
follow the changes in visualization as animations such as the introduction of new edges and
smooth movement of nodes in a force-directed graph. Interactive version of this graph can
be found at http://ahmetrasit.com/blosum/. User can move the mouse over an amino acid
and get a  detailed information about its  molecular weight, frequency, codons, a brief
information and chemical representation of the molecule at the  bottom-right corner
of the page. The user can also change BLOSUM identity value using the slider on the
top-left, change substitution score threshold to change the network structure using
the second slider on the top,  change node color to represent different chemical properties
*ahmet.ozturk@metu.edu.tr
  Proceedings of CELLmicrocosmos neXt 2014 (DOI: 10.2390/biecoll-next2014-5)  13
Figure 1: Network representations of BLOSUM62 (left) and BLOSUM45 (right) matrices.
of  amino  acids  (horizontal  bars  on  the  center-right  side  is  the color  map of  selected
chemical  properties),  and  lastly,  change  node  size  to  represent  different  numerical
properties of each amino acid according to the categories listed on the top-right side.
Figure  2:  Screenshot  of  the  Interactive  BLOSUM  Graph  GUI  available  at
http://ahmetrasit.com/blosum/ 
1.2 Content of the Workshop
In this talk, I will give some examples of how visualization helps us understand the story
behind  the  data  and  I  will  demonstrate  how to  manipulate  HTML elements  using  D3.
Finally,  I  will  introduce  basic  interactive  graphs  created  with  D3  and  show  how  to
manipulate a single graph layout with different data.
References
[1] D.  Skau.  Why  D3.js  is  So  Great  for  Data  Visualization  |  Visually  Blog.  [Online].
Available: http://blog.visual.ly/why-d3-js-is-so-great-for-data-visualization. [Accessed:
17-Sep-2014].
[2] M.  Bostock.  D3.js  -  Data-Driven  Documents.  [Online].  Available:  http://d3js.org.
[Accessed: 13-Sep-2014].
[3] nsonnad.  Object  constancy  with  multiple  sets  of  time-series.  [Online].  Available:
http://bl.ocks.org/nsonnad/4175202. [Accessed: 17-Sep-2014].


German

Web-based Visualization with D3

Vompras, Johanna

Einsatz von Drupal in Forschung und Bildung

Pieper, Dirk

E-Books

Gris, Iván

Novick, David

Rivera, Diego A.

Gutiérrez, Mario

 
1 Main Research Themes 
Two related goals of IVA research involve increasing 
the believability and perceived trustworthiness of 
agents and increasing the user's sense of engagement. 
In our research, we use human-human interaction as a 
model for agent behaviors that build rapport between 
humans and agents. Our paralinguistic model of rap-
port (Novick & Gris, 2014) comprises a sense of emo-
tional connection, a sense of mutual understanding, 
and a sense of physical connection. Our research, 
which focuses primarily on the sense of physical con-
nection, seeks to increase the naturalness of non-verbal 
interaction to correspondingly increase human-IVA 
rapport (Tickle-Degnan & Rosenthal, 1987; Huang et 
al., 2011). For these studies, we have developed two 
agents, with variable non-verbal behaviors, in immer-
sive games designed to maintain users’ engagement. 
1.1 Familiarity Agent 
This agent is part of a longer-term project to provide 
IVAs with behaviors that enable them to build and 
maintain rapport with their human partners. We focus 
on paralinguistic behaviors, especially nonverbal be-
haviors, and their role in communicating rapport. The 
Our IVA guides its players through a speech-
controlled game called “Escape from the Castle of the 
Vampire King (see Figure 1), through which we meas-
ure the familiarity between humans and agents across 
two interaction sessions. We studied whether increas-
ing amplitude of nonverbal paralinguistic behaviors 
leads to an increased perception of physical connect-
edness between humans and ECAs (Novick & Gris, 
2013; Gris, Novick, Gutierrez & Rivera, in press). 
1.2 Two-Way Virtual Rapport Agent 
This agent is a work in progress in which we make use 
of full-body gesture recognition to create a sense of 
physical connection between users and agents. To 
achieve this, we guide users through a jungle-island 
survival scenario (see Figure 2). We focus on collabo-
rative physical activities that take place in the virtual 
environment but require the user’s physical action, 
such as lighting a fire or spearing a fish.  
 
 
Figure 1. Interaction with the “Escape from the Castle 
of the Vampire King” agent. 
 
 
Figure 2. Jungle survival agent. 
2 Current Architectures and Standards 
We developed our agents independently of SAIBA, 
BML, and FML, in part because we wanted to under-
stand agent development from the ground up and in 
part because our research requires some less-common 
features, such as recognition of full body gestures, 
blending animations, and agent portability and reuse. 
The project would likely have benefited from using 
PBL as a foundation for gesture recognition. 
 
Our implementation uses Unity 4, a Microsoft Kinect, 
and the Windows Speech SDK, interfaced and net-
 
 
UTEP’s AGENT Architecture 
 
Iván Gris,  David Novick, Diego A. Rivera, Mario Gutiérrez 
The University of Texas at El Paso 
worked with each other and synchronized to handle the 
agent’s complex behavior. Figure 3 presents an im-
plementation-level outline of the system’s components 
and their relationships. We use the Unity 4 game en-
gine to display our agents and Unity’s Mecanim sys-
tem to create an extensive array of animations. 
 
 
Figure 3. Implementation-level architecture of the 
software-hardware interfaces and feature handling. 
 
Animations are played by a state graph that follows 
user-specified parameters of when an animation should 
start, end, or blend with another animation. Multiple 
animations can be blended to obtain a completely dif-
ferent animation in real time and give the user the im-
pression that the agent never moves in exactly the 
same way twice. Animations are divided layers that 
can control different parts of the body, so multiple 
animations can be played at the same time and affect 
different limbs of the agent. The animations are played 
when the system decodes a message sent by the dialog 
tree that has the information about the specific anima-
tion to be played, the length of the dialog that the agent 
will say, and the position where the agent and the 
player should be. 
 
To describe interactive scenes, we developed a dialog 
interpreter that parses an XML document and links 
dialog states through conditionals. The file includes 
the responses anticipated by the systems at relevant 
parts of the scene. After the interpreter compiles the 
file, it builds a dialog tree that contains the relation-
ships of the dialogs segments through the storyline.  
3 Future of Architectures and Standards for 
IVAs 
Current generations of users are accustomed to hyper-
realistic videogame characters and movie assets. A key 
aspect of these advanced-gen characters is the fluid 
and realistic way in which they move, which is usually 
handmade through motion capture. IVAs are a step 
behind, as their movement is generated on the fly. We 
would like to see a standard that enables research 
groups to exchange animation sets or subsets (anima-
tion of specific joints) that can be blended at a later 
time to create unlimited specialized movements.’ 
 
Another key feature of IVAs is the real-time percep-
tion of a user’s full-body gestures. For most IVAs, this 
ability is limited, partly because the necessary setup is 
more elaborate, requiring more than a computer and a 
screen. Now that it is possible to track skeleton joint 
movement, gesture recognition can be more than a set 
of image-analysis algorithms with limited tracking 
capabilities, complicated setups, and long render 
times. We hope that a standard can be set to capture 
gestures or poses based on skeleton data, so that stud-
ied gestures can easily be ported into the recognizers 
and applied across agents.  
 
More broadly, we see a need to leverage standards and 
shared tools through more effective organization of the 
research community. Our experience suggests that 
newcomers to the field would be well served by hav-
ing a single place from which to obtain standards and 
tools rather than having to visit our research groups’ 
individual sites.  
4 Suggestions for Discussion 
Our approach for IVA design is modular. Our agents 
recognize full-body gesture recognition and produce 
nonverbal behaviors using multiple technologies that 
we later synchronize through a network. We are inter-
ested in how to develop these modular technologies, 
how can we share them across research groups, and 
how we can address difficult interaction problems with 
them. Specific questions include: 
 
• How can shared resources (e.g., BML, PML, 
SABIA) support extension to novel modes of in-
teraction? 
• Given the popularity and accessibility of the latest 
gaming technologies, how can we make use of 
them to develop quick, inexpensive and portable 
IVA prototypes?  
• With the increased IVA functionality and interac-
tive capabilities, what requirements and standards 
should we take into consideration when develop-
ing agents that interact not only with humans but 
with other agents at the same time? 
• Can we define a clear separation of IVAs’ archi-
tectural components between their domain and the 
agent’s features (e.g., gesture recognition, speech 
recognition, virtual environments, gesture and 
pose handling, AI elements)? How can we use this 
to create IVAs, and specific features that can be 
shared across research groups and disciplines? 
References 
Gris, I. Adaptive virtual rapport for embodied conver-
sational agents. In Proceedings of the 15th ACM on 
International conference on multimodal interaction. 
ACM, 2013. 
Huang, L., Morency, L.-P., and Gratch, J. Virtual rap-
port 2.0. In Intelligent Virtual Agents, Lecture 
Notes in Computer Science, Vol. 6895, chapter 8, 
pages 68-79. Springer, Berlin, Heidelberg, 2011. 
Novick, D., and Gris, I. Building rapport between hu-
man and ECA: a pilot study. In Proceedings of HCI 
International 2014, Heraklion, Greece, July, 2014, 
Lecture Notes in Computer Science, vol. 8511, 472-
480. 
Gris I., Novick, D., Rivera D.A., Gutierrez, M. Rec-
orded speech, virtual environments, and the effec-
tiveness of embodied conversational agents. In 
Proceedings of Intelligent Virtual Agents, 2014, in 
press. 
Tickle-Degnen, L., and Rosenthal, R. Group rapport 
and nonverbal behavior. In Group processes and in-
tergroup relations, pages 113-136. Sage, Newbury 
Park, CA, 1987.  
  
Biographical Sketches 
 
Ivan Gris is a Ph.D. student at the 
University of Texas at El Paso 
working under the supervision of 
Dr. David Novick. He works in the 
Interactive Systems Group as part 
of the Advanced aGent ENgage-
ment Team as the project manager 
for developing full body embodied 
conversational agents and immer-
sive, interactive environments. When he is not work-
ing on his dissertation, he is working on the research 
and development for two start-up companies he creat-
ed. One of these companies is a tech start-up that uses 
embodied conversational agents, animatronics, scene 
development, and visual and special effects to develop 
a unique role-playing themed experience.  
 
Diego A. Rivera is an undergradu-
ate student majoring in Computer 
Science. He works with the Ad-
vanced aGent ENgagement Team 
as a lead animator. He works inte-
grating and developing the systems 
that handle the animations for the 
agents using Unity 3D. In addition 
he is the main programmer of the 
dialog interpreter and several other tools and pipelines 
that control the agent’s behavior. Recently Diego ob-
tained an internship on the Institute for Creative Tech-
nologies at the University of South-California where 
he is working under the supervision of Dr. Chad Lane 
as an integration programmer for Virtual Human and 
Intelligent Tutoring Research. 
 
 Mario Gutierrez is a Master’s 
student at the University of Texas 
at El Paso. Mario is the lead 
programmer for the Advanced 
aGent Engagement Team. He is 
currently working on his Master’s 
project, which consists of 
implementing a memory 
knowledge base for an ECA. He 
aims to develop agents with memory of previous 
interactions and dialogs with the users, and then recall 
them at the appropriate time and in a correct maner. 
 
David Novick is the Mike 
Loya Distinguished Chair and 
Professor of Computer Science 
at the University of Texas at El 
Paso. He earned his J.D. at Har-
vard University in 1977 and his 
Ph.D. in Computer and Infor-
mation Science at the Universi-
ty of Oregon in 1988. His 
research has included work on turn-taking and gaze in 
artificial agents. He built UTEP’s Advanced aGent 
ENgagement Team, whose research focuses on inter-
active systems and, especially, building rapport in 
multimodal conversation. 
 


UTEP's AGENT Architecture

Wallmeyer, Thorben

Sommer, Björn

Kormeier, Benjamin

  Proceedings of CELLmicrocosmos neXt 2014 (DOI: 10.2390/biecoll-next2014-8)  18
Towards Pathway Prediction and Subcellular Localization by
using FraMeTex
Thorben Wallmeyer*, Björn Sommer, Benjamin Kormeier
Bio-/Medical Informatics Department, Bielefeld University/Center for Biotechnology
Universitätsstr. 25, 33615 Bielefeld, Germany
1 Introduction
The  visualization  and  localization  of  biochemical  networks  represent  major  tasks  of
bioinformatics. In the future, they could be used as additional criteria to find potential
-omics interaction partners. Moreover, this visualization and localization can help scientists
to find appropriate experiments or to create detailed problem-oriented pathways. Today,
we can use database integration and text mining methods for the prediction of biological
networks. However, we can use the same applications for the localization prediction of
biological networks and their components [1]. Using this information, we are able to model
and visualize biochemical networks in 2D or 3D. For this purpose, we use our database
integration infrastructure BioDWH with the DAWIS-M.D., as well as ANDCell as the base for
the subcellular localization of the CELLmicrocosmos 4.2 PathwayIntegration (CmPI) project
[1–4]. Here, we discuss how this approach can be extended by using the new FraMeTex text
mining framework.
2 Data integration
High-throughput methods generate, in short time, data concerning the whole genome of an
organism.  Hence,  enormous,  heterogeneous  and  versatile  data  is  produced.  The  total
number of databases, as well the data itself, is continuously increasing. Data distribution
and  heterogeneity  causes  big  problems  in  biological  data  integration.  Therefore,  we
introduced a bioinformatics data warehouse information system DAWIS-M.D. that can be
accessed with the CmPI for tailor-made 3D cell and pathway visualization. DAWIS-M.D. is a
platform-independent data warehouse approach for metabolic data that is based on the
BioDWH data warehouse infrastructure.  The information system combines a number  of
relevant biological databases. For the subcellular localization, BRENDA, Gene Ontology,
Reactome, and UniProt are used. The content of the data warehouse is divided into 12
diverse domains, which can be queried via a web-bases graphical user interface that can
be accessed with any common web-browser. The DAWIS-M.D. application provides search
forms for the following domains: Compound, Disease, Drug, Transcription Factor, Enzyme,
Gene, Glycan, Gene Ontology, Pathway, Protein, Reaction and Reaction Pair. Moreover, it is
possible to identify relationships and interactions spanning multiple biological domains [3].
DAWIS-M.D.  provides  also  a  web-service  to  the  CmPI.  Networks  can  be  localized  and
displayed  in  a  cell  environment,  edited  and  extended  by  using  the  CmPI  software
application [1].
3 Text mining
The number of  text  mining algorithms that can be helpful  in  reconstructing biological
networks from text data is unclear. Therefore, selecting and using the best text mining
algorithm  for  a  specific  task  is  challenging.  In  addition,  the  heterogeneity  of  their
interfaces and data structures for representing extracted facts is even complicating their
*thorben.wallmeyer@googlemail.com
  Proceedings of CELLmicrocosmos neXt 2014 (DOI: 10.2390/biecoll-next2014-8)  19
application. We addressed this problem by developing a powerful framework. FraMeTex
(Framework  for  Medical  Text  mining  and  knowledge  engineering)  offers  a  highly
customizable analysis pipeline that is able to consult different text mining algorithms for
knowledge  extraction.  Furthermore,  it  supports  the  adaption  of  text  sources  to  be
analyzed as  well  as  the filtering of  relevant  data.  Extracted knowledge like  biological
pathways or localizations can afterwards be persisted within a deductive knowledgebase.
By using additional rules it is able to identify additional relations. The offered analysis
pipeline of FraMeTex is divided into several logical modules that provide a convenient API
(see figure). It allows a seamless and flexible integration of its functional modules within
almost  any  application.  Each  module  can  be  used  standalone  or  chained  with  other
modules  to  build  a  complex  workflow.  We  developed  a  workflow  that  filters  Medline
abstracts by keywords and analyzes them with two different text mining algorithms. This
enables us to extract reliable pathway and localization data from Medline. GENIAs Named
Entity  Recognition  (NER)  web  service  is  responsible  for  tagging  potential  biomedical
entities within previously selected Medline abstracts [5]. It distinguishes between proteins,
cell lines, cell types, DNA and RNA terms. Additionally Enju performs a semantic parsing of
the abstracts and returns Predicate Argument Structures (PAS). The analysis results of both
algorithms  are  finally  merged.  PAS  that  deal  with  biomedical  entities  are  most  likely
supposed to describe biological pathways or localizations.
4 Conclusion and Outlook
The FrameTex localization should
be  used  to  try  to  extend  and
verify  the  knowledge  of  CmPI.
We  will  do  this  by  using  the
described and a newly developed
workflow.  The  additional
workflow will  use different text
mining  algorithms  which  allows
an  additional  cross  check  of
results.  Especially  the  critical
NER  process  may  be  performed
better by other algorithms.
Figure 1: The FraMeTex pipeline
5 References
[1] B. Sommer, B. Kormeier, P. S. Demenkov, P. Arrigo, K. Hippe, Ö. Ates, A. V. Kochetov, V.
A. Ivanisenko, N. A. Kolchanov, and R. Hofestädt. Subcellular Localization Charts: A
new visual  methodology for the semi-automatic localization of protein-related data
sets. Journal of Bioinformatics and Computational Biology, 11(1):1340005, 2013.
[2] T. Töpel, B. Kormeier, A. Klassen, and R. Hofestädt. BioDWH: a data warehouse kit for
life science data integration. Journal of Integrative Bioinformatics, 5(2):93, 2008.
[3] K.  Hippe,  B.  Kormeier,  T.  Töpel,  S.  Janowski,  and  R.  Hofestädt.  DAWIS-MD-a  data
warehouse system for metabolic data. GI Jahrestagung, 2:720–725, 2010.
[4] O. A. Podkolodnaya, E. E. Yarkova, P. S. Demenkov, O. S. Konovalova, V. A. Ivanisenko,
and N. A. Kolchanov. Application of the ANDCell computer system to reconstruction and
analysis of associative networks describing potential relationships between myopia and
glaucoma. Russian Journal of Genetics: Applied Research, 1(1):21–28, 2011.
[5] J.-D. Kim, T. Ohta, Y. Tateisi, and J. Tsujii. GENIA corpus—a semantically annotated 
corpus for bio-textmining. Bioinformatics, 19(suppl 1):i180–i182, 2003.


Towards Pathway Prediction and Subcellular Localization by using FraMeTex

Jahn, Najko

Kerndatensatz Forschung

Pietsch, Christian

Peil, Vitali

Langzeitarchivierung mit LOCKSS und Zeitreisen mit Memento

Summann, Friedrich

Katalog.plus! Die integrierte Suchumgebung der Universitätsbibliothek Bielefeld

Wolf, Sebastian

Grundkenntnisse der Wikipedia-Mitarbeit

Veselkov, Kirill A.

  Proceedings of CELLmicrocosmos neXt 2014 (DOI: 10.2390/biecoll-next2014-4)  11
Top-down systems biology approaches for
“omics”-based tissue and biofluid analytics
Kirill A. Veselkov*
Faculty of Medicine, Department of Surgery & Cancer,
Imperial College London, London SW7 2AZ, United Kingdom
Top-down systems biology identifies molecular interaction networks in health and disease
using  system-wide  data  generated  by  high-throughput  –omics  technologies.  The  rapid
advances in  the post-genomic analytical  technologies,  most  notably  Mass  Spectrometry
(MS), has made possible to simultaneously analyze hundreds to thousands of metabolites in
biological fluids and intact tissues at a relatively low cost. The ability to generate, model,
and interpret such metabolic data in relation to benchmark patient clinical information,
enabled by powerful  bioinformatics  platforms,  makes  personalized medicine a  tangible
proposition.  The  advancements  in  the  field  of  metabolic  phenotyping  have  focused
primarily on instrument development and validation of the MS technologies. 
However, despite the significant advances in these areas, the major barrier to its wider
adoption and transition into main stream medicine and industry has been the challenge of
interpreting  and  managing  the  huge  data  sets  MS  generates.  The  interpretation  of
metabolic  data  is  dependent  on  the  advances  made  by  reconstruction  of  molecular
networks via cellular bottom-up systems biology models. 
This presentation outlines innovative bioinformatics solutions for emerging technologies of
mass  spectrometry  imaging  (MSI)  and  rapid  evaporative  ionization  mass  spectrometry
(REIMS),  targeting  unmet  medical  needs  of  cancer  network-driven  diagnosis  and  rapid
pathogen identification in sepsis. The translational bioinformatics platforms are presented
that allow intuitive clinically-oriented interrogation of MS datasets in an integrated and
streamlined  fashion.  These  developments  aim  to  overcome  current  methodological
roadblocks for effective clinical translation of MS technologies since the existing workflows
for data processing rely on a heterogeneous array of bioinformatic packages which are
poorly optimized and only suitable for specialists in the field [1], [2].
References
 
[1] N. Abbassi-Ghadi, K. Veselkov, S. Kumar, J. Huang, E. Jones, N. Strittmatter, H. Kudo,
R. Goldin, Z. Takáts, and G. B. Hanna. Discrimination of lymph node metastases using
desorption  electrospray  ionisation-mass  spectrometry  imaging.  Chemical
Communications, 50(28):3661–3664, 2014.
[2] K. A. Veselkov, R. Mirnezami, N. Strittmatter, R. D. Goldin, J. Kinross, A. V. Speller, T.
Abramov, E. A. Jones, A. Darzi, and E. Holmes. Chemo-informatic strategy for imaging
mass  spectrometry-based  hyperspectral  profiling  of  lipid  signatures  in  colorectal
cancer. Proceedings of the National Academy of Sciences, 111(3):1216–1221, 2014.
*kirill.veselkov04@imperial.ac.uk


Top-down systems biology approaches for 'omics'-based tissue and biofluid analytics

Not a member yet