• Aucun résultat trouvé

Intelligent Algorithms in Ambient and Biomedical Computing

N/A
N/A
Protected

Academic year: 2022

Partager "Intelligent Algorithms in Ambient and Biomedical Computing"

Copied!
329
0
0

Texte intégral

(1)
(2)

Intelligent Algorithms in Ambient and Biomedical Computing

(3)

Philips Research

Editor-in-Chief Dr. Frank Toolenaar

Philips Research Laboratories, Eindhoven, The Netherlands

SCOPE TO THE ‘PHILIPS RESEARCH BOOK SERIES’

As one of the largest private sector research establishments in the world, Philips Research is shaping the future with technology inventions that meet peoples’ needs and desires in the digital age. While the ultimate user benefits of these inventions end up on the high-street shelves, the often pioneering scientific and technological basis usually remains less visible.

This ‘Philips Research Book Series’ has been set up as a way for Philips researchers to contribute to the scientific community by publishing their comprehensive results and theories in book form.

Dr. Rick Harwig

VOLUME 7

(4)

Edited by

and

Intelligent Algorithms in Ambient and Biomedical Computing

Jan Korst

Philips Research Laboratories, Eindhoven, The Netherlands

Philips Research Laboratories, Eindhoven, The Netherlands

Emile Aarts

The Netherlands

Philips Research Laboratories, Eindhoven,

Wim Verhaegh

(5)

Published by Springer,

P.O. Box 17, 3300 AA Dordrecht, The Netherlands.

Printed on acid-free paper

All Rights Reserved

© 2006 Springer

No part of this work may be reproduced, stored in a retrieval system, or transmitted in any form or by any means, electronic, mechanical, photocopying, microfilming,

recording or otherwise, without written permission from the Publisher, with the exception of any material supplied specifically for the purpose of

being entered and executed on a computer system, for exclusive use by the purchaser of the work.

Printed in the Netherlands.

www.springer.com

A C.I.P. Catalogue record for this book is available from the Library of Congress.

ISBN-10 1-4020-4953-6 (HB) ISBN-10 1-4020-4995-1 (e-book) ISBN-13 978-1-4020-4953-8 (HB) ISBN-13 978-1-4020-4995-8 (e-book)

(6)

Contents

Contributing Authors xi

Preface xvii

Acknowledgments Part I Healthcare

1. Bioscience Computing and the Role of Computational Simulation in

3 Christopher D. Clack

1.1. Introduction to bioscience computing 3

1.2. Simulating adaptive behaviour 8

1.3. Impact and future directions for bioscience computing 15

1.4. Summary and conclusions 16

References 17

2. The Many Strands of DNA Computing 21

Nevenka Dimitrova

2.1. Introduction 21

2.2. DNA computing 22

2.3. Synthetic biology 31

2.4. Conclusion and future directions 33

References 34

3. Bio-Inspired Data Management 37

3.1. Introduction 37

3.2. Data cell overview 40

3.3. The communication infrastructure 46

3.4. The life cycle 49

3.5. Application challenges 51

3.6. Conclusion 54

References 55

Biology

Martin L. Kersten andArnoP.J.M. Siebes

xxi

(7)

vi Contents

4.2. Biological consciousness 59

4.3. Machine consciousness 60

4.4. Is it relevant? 65

4.5. Applications 67

4.6. Conclusion 69

References 69

Part II Lifestyle

5. Optimal Selection of TV Shows for Watching and Recording 73 Wim F.J. Verhaegh

5.1. Introduction 73

5.2. Problem definition 75

5.3. Computational complexity 76

5.4. Scheduling shows for watching 77

5.5. A dynamic programming approach 78

5.6. Run time improvements 80

5.7. Experiments 82

5.8. Conclusion 85

References

6. Movie-in-a-Minute: Automatically Generated Video Previews 89 Mauro Barbieri, Nevenka Dimitrova and Lalitha Agnihotri

6.1. Introduction 89

6.2. Related work 90

6.3. Requirements 91

6.4. Formal model 92

6.5. Implementation and results 95

6.6. Need for personalization 99

6.7. Conclusions 100

References 100

7. Features for Audio Classification: Percussiveness of Sounds 103 Janto Skowronek and Martin McKinney

7.1. Introduction 103

7.2. Feature extraction algorithm 105

7.3. Experiments 109

7.4. Summary 117

References 117

4. An Introduction to Machine Consciousness 57

Kees van Zon

4.1. Introduction 57

87

(8)

Contents vii

9. Approximate Semantic Matching of Music Classes on the Internet 133

9.1. Introduction 133

9.2. Semantic coordination 135

9.3. Internet music schemas 136

9.4. Approximate matching 139

9.5. Experiment with approximate matching 140

9.6. Future work 144

9.7. Conclusion 146

References 146

10. Ontology-Based Information Extraction from the World Wide Web 149

10.1. Introduction 149

10.2. Problem definition 151

10.3. Solution approach 153

10.4. Case study: Finding famous people on the Web 154

10.5. Concluding remarks 165

References 165

11. Privacy Protection in Collaborative Filtering by Encrypted Computation

169 Wim F.J. Verhaegh, Aukje E.M. van Duijnhoven, Pim Tuyls

and Jan Korst

11.1. Introduction 169

11.2. Memory-based collaborative filtering 171

11.3. Encryption 176

11.4. Encrypted user-based algorithm 178

11.5. Encrypted item-based algorithm 182

11.6. Conclusion 183

References 184

8. Extracting the Key from Music 119

Steffen Pauws

8.1. Introduction 119

8.2. Musical pitch and key 121

8.3. Method 122

8.4. Evaluation 129

8.5. Conclusion 130

References 131

Zharko Aleksovski, Warner ten Kate and Frank van Harmelen

Jan Korst, Gijs Geleijnse, Nick de Jong and Michael Verschoor

(9)

viii Contents

13.1. Introduction 215

13.2. RDF graphs and simple entailment 219

13.3. RDFS entailment andD* entailment 223

13.4. pD* entailment 233

13.5. Conclusion 241

References 241

243 and Nikos Vlassis

14.1. Introduction 243

14.2. Bayesian networks for dynamic systems analysis 244

14.3. Localization of a mobile platform 250

14.4. Tracking with distributed cameras 253

14.5. Conclusions and remaining issues 257

References 258

15. Private Profile Matching 259

Berry Schoenmakers and Pim Tuyls

15.1. Introduction 259

15.2. Preliminaries 261

15.3. Secure approximate matching w.r.t. Hamming distance 267

15.4. Conclusion 271

References 272

13. Semantic Web Ontologies and Entailment: Complexity Aspects 215 Herman J. ter Horst

Part III Technology

12. A First Look at the Minimum Description Length Principle 187 Peter D. Gr¨unwald

12.1. Introduction and overview 187

12.2. The fundamental idea: Learning as data compression 189

12.3. MDL and model selection 192

12.4. Crude and refined MDL 195

12.5. The MDL philosophy 202

12.6. MDL and Occam’s razor 205

12.7. History 207

12.8. Challenges for MDL: The road ahead 208

12.9. Summary, conclusion and further reading 210

References 211

Wojciech Zajdel, Ben J.A. Krose

14. Bayesian Methods for Tracking and Localization

¨

(10)

Contents ix

17.9. Conclusions 314

References 315

Index 317

16. Air Fair Scheduling for Multimedia Transmission over Multi-Rate Wireless LANs

273 and Kang G. Shin

16.1. Introduction 273

16.2. Fairness in wireless/mobile networks 277

16.3. AFS in an IEEE 802.11e wireless LAN 282

16.4. Station scheduler 285

16.5. Local scheduler (LS) 287

16.6. Numerical and simulation results 291

16.7. Experimental setup and results 293

16.8. Conclusions 296

References 296

17. High Throughput and Low Power Reed Solomon Decoder for Ultra Wide Band

299 Akash Kumar and Sergei Sawitzki

17.1. Motivation 299

17.2. Introduction to Reed Solomon 300

17.3. Channel model 301

17.4. Architecture design options 304

17.5. Design flow 308

17.6. Results 309

17.7. Benchmarking 312

17.8. Optimisations to design 312

Sai Shankar N., Richard Y. Chen, Ruediger Schmitt, Chun-Ting Chou

(11)

Contributing Authors

Emile Aarts

Philips Research Laboratories Eindhoven

High Tech Campus 36, 5656 AE Eindhoven, The Netherlands emile.aarts@philips.com

Lalitha Agnihotri

Philips Research Laboratories USA

345 Scarborough Rd., Briarcliff Manor, NY 10510, USA lalitha.agnihotri@philips.com

Zharko Aleksovski

Philips Research Laboratories Eindhoven

High Tech Campus 31, 5656 AE Eindhoven, The Netherlands zharko@cs.vu.nl

Mauro Barbieri

Philips Research Laboratories Eindhoven

High Tech Campus 34, 5656 AE Eindhoven, The Netherlands mauro.barbieri@philips.com

Richard Y. Chen

Philips Research Laboratories USA

345 Scarborough Rd., Briarcliff Manor, NY 10510, USA richard.chen@philips.com

Chun-Ting Chou University of Michigan

1301 Beal Avenue, Ann Arbor, MI 48109, USA choujt@umich.edu

xi

(12)

xii Contributing Authors Christopher D. Clack

University College London

Gower Street, London WC1E 6BT, United Kingdom clack@cs.ucl.ac.uk

Nevenka Dimitrova

Philips Research Laboratories USA

345 Scarborough Rd., Briarcliff Manor, NY 10510, USA nevenka.dimitrova@philips.com

Aukje E.M. van Duijnhoven Numerando Groep

Amersfoortseweg 10e, 3705 GJ Zeist, The Netherlands aukjevanduijnhoven@hotmail.com

Gijs Geleijnse

Philips Research Laboratories Eindhoven

High Tech Campus 34, 5656 AE Eindhoven, The Netherlands gijs.geleijnse@philips.com

Frank van Harmelen Vrije Universiteit Amsterdam

De Boelelaan 1081a, 1081 HV Amsterdam, The Netherlands frank.van.harmelen@cs.vu.nl

Herman J. ter Horst

Philips Research Laboratories Eindhoven

High Tech Campus 31, 5656 AE Eindhoven, The Netherlands herman.ter.horst@philips.com

Nick de Jong Tiobe Software B.V.

De Zaale 11, 5612 AJ Eindhoven, The Netherlands nick.de.jong@tiobe.com

Warner ten Kate

Philips Research Laboratories Eindhoven

High Tech Campus 31, 5656 AE Eindhoven, The Netherlands warner.ten.kate@philips.com

(13)

Contributing Authors xiii Martin L. Kersten

Centrum voor Wiskunde en Informatica

Kruislaan 413, 1098 SJ Amsterdam, The Netherlands martin.kersten@cwi.nl

Jan Korst

Philips Research Laboratories Eindhoven

High Tech Campus 34, 5656 AE Eindhoven, The Netherlands jan.korst@philips.com

Ben J.A. Kr¨ose

University of Amsterdam

Kruislaan 403, 1098 SJ Amsterdam, The Netherlands krose@science.uva.nl

Akash Kumar

Philips Research Laboratories Eindhoven

High Tech Campus 45, 5656 AE Eindhoven, The Netherlands akakumar@natlab.research.philips.com

Martin McKinney

Philips Research Laboratories Eindhoven

High Tech Campus 36, 5656 AE Eindhoven, The Netherlands martin.mckinney@philips.com

Steffen Pauws

Philips Research Laboratories Eindhoven

High Tech Campus 34, 5656 AE Eindhoven, The Netherlands steffen.pauws@philips.com

Sai Shankar N.

Qualcomm Standards Engineering Dept.

5775 Morehouse Drive, San Diego, CA 92121, USA nsai@qualcomm.com

Sergei Sawitzki

Philips Research Laboratories Eindhoven

High Tech Campus 31, 5656 AE Eindhoven, The Netherlands sergei.sawitzki@philips.com

(14)

xiv Contributing Authors Ruediger Schmitt

Philips Research Laboratories USA

345 Scarborough Rd., Briarcliff Manor, NY 10510, USA ruediger.schmitt@philips.com

Berry Schoenmakers

Technische Universiteit Eindhoven

Den Dolech 2, 5612 AZ Eindhoven, The Netherlands berry@win.tue.nl

Kang G. Shin

University of Michigan

1301 Beal Avenue, Ann Arbor, MI 48109, USA kgshin@umich.edu

Arno P.J.M. Siebes Universiteit Utrecht

Padualaan 14, 3584 CH Utrecht, The Netherlands arno.siebes@cs.uu.nl

Janto Skowronek

Philips Research Laboratories Eindhoven

High Tech Campus 36, 5656 AE Eindhoven, The Netherlands janto.skowronek@philips.com

Pim Tuyls

Philips Research Laboratories Eindhoven

High Tech Campus 34, 5656 AE Eindhoven, The Netherlands pim.tuyls@philips.com

Wim F.J. Verhaegh

Philips Research Laboratories Eindhoven

High Tech Campus 34, 5656 AE Eindhoven, The Netherlands wim.verhaegh@philips.com

Michael Verschoor

Technische Universiteit Eindhoven

Den Dolech 2, 5612 AZ Eindhoven, The Netherlands m.p.f.verschoor@tm.tue.nl

(15)

Contributing Authors xv Nikos Vlassis

University of Amsterdam

Kruislaan 403, 1098 SJ Amsterdam, The Netherlands vlassis@science.uva.nl

Wojciech Zajdel

University of Amsterdam

Kruislaan 403, 1098 SJ Amsterdam, The Netherlands wzajdel@science.uva.nl

Kees van Zon

Philips Research Laboratories USA

345 Scarborough Rd., Briarcliff Manor, NY 10510, USA kees.van.zon@philips.com

(16)

Preface

The rapid growth in electronic systems in the past decade has boosted re- search in the area of computational intelligence. As it has become increasingly easy to generate, collect, transport, process, and store huge amounts of data, the role of intelligent algorithms has become prominent in order to visualize, manipulate, retrieve, and interpret the data. For instance, intelligent search techniques have been developed to search for relevant items in huge collec- tions of web pages, and data mining and interpretation techniques play a very important role in making sense out of huge amounts of biomolecular measure- ments. As a result, the added value of many modern systems is no longer determined by hardware only, but increasingly by the intelligent software that supports and facilitates the user in realizing his or her objectives.

Over the past years, considerable progress has been made in the area of com- putational intelligence, which can be positioned at the intersection of computer science, discrete mathematics, and cognitive science. This has led to a grow- ing community of practitioners within Philips Research that develop, analyze, and apply intelligent algorithms. The Symposium on Intelligent Algorithms (SOIA) intends to provide this community of practitioners with a platform to exchange information. The first edition of SOIA, held in 2002, addressed the topic of intelligent algorithms in ambient intelligence. To share the output of the symposium with a larger audience, a selection of papers was edited and published by Kluwer in the Philips Research Book Series under the title “Al- gorithms in Ambient Intelligence.” For the second edition, held in 2004, the scope of the symposium was broadened so as to comply with the three main

consists of 17 chapters, divided over three parts corresponding to the strategic topics mentioned above. The main topic in Healthcare is the understanding of biological processes, for Lifestyle the main topic is content retrieval and manipulation, and finally for Technology most contributions relate to media processing. Below we present more detailed information about the individual chapters.

logy. Again a selection of papers was edited, resulting in the present book. It topics of the Philips company strategy, i.e., Healthcare, Lifestyle and Techno-

xvii

(17)

xviii Preface Part I consists of four chapters. In Chapter 1, Chris Clack discusses the topic of modeling biological systems, thus allowing to perform in-silico exper- iments by means of computer simulation, to formulate hypotheses. In Chapter 2, Nevenka Dimitrova gives an overview of the reverse approach, where one does not use computers to simulate biological processes, but where one uses biology to perform computations, in DNA computing and synthetic biology.

In Chapter 3, Martin Kersten and Arno Siebes discuss data management in- spired by biology, resulting in an organic database system. In Chapter 4, Kees van Zon discusses how to achieve machine consciousness, and how it can be applied.

problem of making a schedule of preferred TV programs, while at the same time selecting TV programs for recording, under the assumption of a limited number of tuners. In Chapter 6, Mauro Barbieri, Nevenka Dimitrova, and Lalitha Agnihotri present a technique to automatically summarize video into a condensed preview, allowing one to quickly browse and access large amounts of stored programs. Chapters 7–9 concerns audio applications. First, Janto Skowronek and Martin McKinney discuss in Chapter 7 the topic of automatic classification of audio and music, for which they developed the automatic ex- traction of the higher-level feature of percussiveness. In Chapter 8, Steffen Pauws presents a technique to automatically extract the key from a piece of music, providing an emotional connotation to it, and making it possible to build well-sounding music mixes. In Chapter 9, Zharko Aleksovski, Warner ten Kate, and Frank van Harmelen address the problem of combining multiple databases of music data in a semantic way, by approximating matches of music classes. Next, Jan Korst, Gijs Geleijnse, Nick de Jong, and Michael Verschoor discuss in Chapter 10 the possibilities to fill a knowledge database, using an ontology to collect and structure data from web pages. In the last chapter of part II, which Wim Verhaegh, Aukje van Duijnhoven, Pim Tuyls, and Jan Ko- rst resolve the privacy issue of population-based recommenders by encrypting the users’ profiles and performing the required algorithms on encrypted data.

Part III consists of six chapters, focusing on the technology underlying in- telligent algorithms and intelligent systems. The first two chapters discuss theoretical aspects of intelligent algorithms. In Chapter 12, Peter Gr¨unwald gives an overview on the minimum description length principle to resolve the problem of model selection, based on the fundamental idea to see learning as a form of data compression. In Chapter 13, Herman ter Horst discusses the computational complexity of reasoning with semantic web ontologies, such as RDF Schema and OWL. Next, Wojciech Zajdel, Ben Kr¨ose, and Nikos Vlas- sis present in Chapter 14 an introduction to dynamic Bayesian networks, and show their application in robot localization and multiple-person tracking. In content management and retrieval. In Chapter 5, Wim Verhaegh discusses the Par tII consists of eight chapters, addressing problems from the area of

(18)

Preface xix Chapter 15, Berry Schoenmaker and Pim Tuyls discuss efficient protocols for securely matching two user profiles, without leaking information on the de- tails of the profiles. Finally, Chapters 16 and 17 address resource issues in intelligent systems. In Chapter 16, Sai Shankar N., Richard Chen, Ruediger Schmitt, Chun-Ting Chou, and Kang Shin revisit fairness in multi-rate wire- less networks, and present a solution to fairly schedule airtime. Finally, in Chapter 17, Akash Kumar and Sergei Sawitzki discuss the design alternatives of Reed Solomon decoders, and address the problem of making optimal design decisions to obtain a high-throughput, low-power solution.

We are convinced that the chapters presented in this book comprise an in- teresting collection of examples of the use of intelligent algorithms in different settings, and that the book reconfirms that the area of computational intelli- gence is a truly challenging field of research.

WIMF.J. VERHAEGH, EMILEAARTS,ANDJANKORST

Philips Research Laboratories Eindhoven

(19)

Acknowledgments

We would like to thank the following people who helped us to review the contributed chapters: Dee Denteneer, Nevenka Dimitrova, Hans van Gagel- donk, Srinivas Gutta, Herman ter Horst, Jan Nesvadba, Dave Schaffer, and Peter van der Stok.

xxi

(20)

Part I

HEALTHCARE

(21)

Chapter 1

BIOSCIENCE COMPUTING AND THE ROLE OF COMPUTATIONAL SIMULATION IN BIOLOGY

Christopher D. Clack

Abstract Bioscience computing exploits the synergy of challenges facing both computer science and biology, drawing inspiration from biology to solve computer sci- ence challenges and simultaneously using new bio-inspired adaptive software to model and simulate biological systems. This chapter first provides an introduc- tion to bioscience computing — discussing the role of computational simulation in terms of hypothesis formulation and prototyping for biologists and medics, and explaining how bioscience computing is both timely and well-suited to sys- tems biology. A concrete example of computational simulation is then provided

— the artificial cytoskeleton, which utilises swarm agents and a cellular au- tomaton to model cell morphogenesis. Morphological adaptation for tasks such as chemotaxis and phagocytosis are presented, and the role of the artificial cy- toskeleton and its swarm-based techniques in both computer science and biology is explained.

Keywords Bioscience computing, systems biology, computational simulation, morphogen- esis, adaptive systems, agent based modelling, swarm agents.

1.1 Introduction to bioscience computing

Bioscience computing exploits the synergy of challenges facing both com- puter science and biology, drawing inspiration from biology to solve problems in computer science and simultaneously using new bio-inspired adaptive soft- ware to model and simulate self-organising, adaptive, biological systems.

There has recently been a substantial increase in inter-disciplinary research interactions between computer science and the life sciences. From the biolo- gist’s perspective, the post-genomic era is characterised by huge amounts of data but little understanding of how genes map to physiological functions, and there is an urgent need for the application of intelligent computing techniques to gain increased understanding. From the computer scientist’s perspective, the new biological data and expanding understanding of biological processes

3

© 2006 Springer. Printed in the Netherlands.

Wim F.J. Verhaegh et al. (Eds.), Intelligent Algorithms in Ambient and Biomedical Computing, 3-19.

(22)

4

provide both an excellent driver for new methods in bioinformatics and an in- creasing source of ideas for new computational techniques in areas such as intelligent systems and artificial life.

The purpose of the first part of this chapter is to provide an introduction to the biological context and to explain the role of bioscience computing within that context.

1.1.1 A change of focus in biology and medicine

The traditional reductionist view of biology is rooted in analysis and bio- physics; it is based on a hierarchical perspective where the functioning of the physiome1 is the deterministic product of a ‘one-way upward causation from genes to cells, organs, system and whole organisms’ [Noble, 2002], and has been remarkably successful with fundamental achievements such as discov- ering the structure of DNA and mapping the genome for not one but several organisms. The traditional role of computer science in biology (e.g. of bioin- formatics) has been to support this endeavour by providing data-handling, data visualisation, numerical simulation and data-mining services.

However, in the post-genomic era the super-abundance of data and relative paucity of understanding, coupled with a clearer perspective of the complexity of living organisms, are causing biologists to question whether the traditional view is sufficient as a basis for a full understanding of nature. The traditional view is giving way to a new biology, often referred to as systems biology.

The rise of systems biology has caused a much closer relationship to develop between biologists and computer scientists. In systems biology, the computer science techniques are no longer merely a data service to the biologists, but are intimately involved in the formulation of biological hypotheses as biologists embrace the process-oriented world of the computer scientist. systems biology considers an organism as a self-organising, adaptive, complex, dynamic sys- tem providing an information framework with global constraints and multiple feedback and regulation paths between high and low levels (e.g. controlling gene expression); the sub-modules are too inextricably connected, there are too many interactions between levels, for a one-way hierarchy to be possible [Noble, 2002]. Biologists now experiment not just in-vivo and in-vitro, but increasingly in-silico. These in-silico experiments are the basis for what we term bioscience computing.

1.1.2 Modelling and simulation

The primary aims of modelling and simulation in biology are to improve understanding of a process or hypothesis, to highlight gaps in knowledge, and

1A glossary of biological terms is provided in Table 1.1 at the end of this chapter.

Christopher D. Clack

(23)

5 to make clear, testable predictions [Kirkwood et al., 2003]. Note, however, that an in-silico experiment itself can never truly be used totest a biological hypothesis — rather, computational simulation in biology should be viewed as a process ofprototyping to assist hypothesis formulation.

Wet-lab experimental techniques tend to focus analytic attention on single mechanisms. By contrast, computational simulation can contribute to the activ- ity of synthesis, of integrating many separate elements that form a network of activity. The resultant interaction and synergy can provide a qualitatively much improved experimental framework. These in-silico results may then guide the choice of (more expensive) subsequent wet-lab experiments.

proaches which generally represent qualitative features of a system, to low- level mechanistic simulations which typically represent quantitative aspects (though abstraction and quantification need not be mutually exclusive con- cepts [Ideker & Lauffenburger, 2003]). Examples of available techniques in- clude statistical data-mining, clustering and classification (e.g. support vector machines), Bayesian networks, Markov chains, fractal theory, Boolean logic, and fuzzy logic. At the mechanistic extreme there are cellular automata and agent-based simulations. Differential equations are widely used and capable of capturing detail at varying levels of abstraction. See Figure 1.1.

Statistical mining

Bayesian networks

Markov chains

Cellular automata (Partial) differential equations

Phenomenological global state

a-priori variables & relations

Mechanistic local state adaptive

Figure 1.1. Comparative spectrum of available techniques.

Phenomenological models tend to focus on theglobalstate of a system. Of- ten they describe an a-priori given set of relations between an a-priori given set of variables [Giavitto et al., 2002]; the two sets cannot evolve jointly with the running system, and very few of these models successfully capture a rich enough semantics to be able to predict complex behaviour [Anderson & Chap- lain, 1998]. By contrast, mechanistic models provide local interaction mod- elling, where cells react (often adaptively) to alocalenvironment, not to the state of the system as a whole (thereby supporting heterogeneity). This leads Techniques. There is a wide spectrum of techniques available to support modelling and simulation, ranging from high-level phenomenological ap-

to a rich model of spatiotemporal dynamics, and offers insights into the Bioscience Computing and Computational Simulation in Biology

(24)

6

Differential equations and partial differential equations provide an excellent mechanism for detailed expression of behaviours of many kinds, but are unsat-

In

An interesting mechanistic approach is the use of cellular automata — e.g.

Scalerandi’s 2D model of cardiac growth dynamics [Scalerandi et al., 2002].

When coupled with agent-based modelling, using a ‘swarm’ of thousands of tiny agents (a mechanism itself inspired by nature) each representing a separate macromolecule, this method has the advantages of both mathematical simplic- ity and that the spatiotemporal fates of individual components (cell, proteins etc.) can be tracked in minute detail. The resulting system is very good at rep- resenting spatiotemporal dynamics and organisational behaviour, particularly for the simulation of adaptive behaviour.

Objects and processes. The specific attraction of computational simulation is that the computational approach corresponds more naturally to the way that biologists think about their subject. Biologists (in particular molecular biolo- gists) naturally focus onobjects,interactionsandprocesses.

Computational simulation permits biologists to express biological systems in terms of computational objects, interactions and processes that relate di- rectly to their biological counterparts and are therefore far easier to under- stand and easier to manipulate than differential equations. Computational simulations can be expressed in terms of information networks and can use interaction-centric models (e.g. local-neighbourhood operations within a cel- lular automaton grid), all of which naturally map onto (for example) cell struc- ture and the interaction of macromolecules.

The experience of systems biology has been that biologists have increas- ingly adopted the computational systems concepts of computer scientists. This should not come as a surprise, since computer scientists have extensive ex- perience of building, modelling, and simulating complex systems that require analysis and synthesis at many different levels of abstraction.

parameters and mechanisms responsible for system dynamics [Gatenby &

Maini, 2003] and for collective organisational behaviour at the microscopic level [Patel, 2004].

McElwain, 2004]. For example, where precise local effects due to inter- isfactory for some highly detailed spatiotemporal behaviours [Araujo &

molecular interactions and random molecular movement are required, a great number of equations must be generated and solved [Succi et al., 2002].

practice, the computational limits on solving a large number of related partial differential equations leads to the technique normally being applied only to abstractionsof internal mechanisms and processes.

Christopher D. Clack

(25)

7

1.1.3 A computational approach to biological complexity

The computational approach to biology enables simulations as dynamic emergent hierarchies of biological complexity, with interactions and feedback between the levels, for example as illustrated in Figure 1.2. At the lowest level, system components are lightweight agents governed by local-neighbourhood rules. The rules provide the system of dynamic interaction between agents, and from this comes the self-organising properties of the simulated organ- ism (threshold parameters may need to be derived via automatic search meth-

neighbourhood rules, the regulatory effects that arise from the self-organising properties of those rules, and sets of global constraints (which may be derived from experimental observation). The result is a complex, dynamic system, which can itself be considered as an agent in a larger network of agents of simi- lar complexity, each undergoing interactions according to local-neighbourhood rules at a higher level, and from which yet more complex behaviour emerges.

co-operative & competitive agent interactions rules of

interaction

emergent regulatory effects

stochastic non-chaotic

patterns of behaviour Higher-Level

Agents

Higher-Level Self Organisation

Higer-Level Emergence

Global Constraints

co-operative & competitive agent interactions rules of

interaction

emergent regulatory effects

stochastic non-chaotic

patterns of behaviour Agents Self Organisation Emergence

Global Constraints Emergent regulatory effects

can constrain behaviour at the

same level or at any other level This dynamic, complex system is itself an agent in a higher-

level system (above)

Figure 1.2. The dynamic emergence of hierarchies of biological complexity.

ods). The emergent behaviour of the system is dependent on a combination of the competitive and co-operative interactions of the underlying local-

hierarchy of levels each can constrain the realisable solutions of the other While emergent behaviour has the potential for chaotic results, in a Bioscience Computing and Computational Simulation in Biology

(26)

8

1.1.4 Summary: The role of bioscience computing

The first part of this chapter has explored the role of bioscience computing in biology and argued that it is both timely and well-suited to the emergence of systems biology: it providesin-silicoexperiments; focuses on interactions and integration of concurrent mechanisms; is intimately involved in the formula- tion of biological hypotheses; manipulates objects, processes and interactions;

is mathematically straightforward, with a low barrier to uptake; and captures rich spatiotemporal detail at low computational cost.

1.2 Simulating adaptive behaviour

This second part provides a concrete example of the bioscience computing techniques discussed in the first part of this chapter, and presents theartificial cytoskeleton, a computational simulation of the development and adaptation of the shape and form of an organism: morphogenesis. The work is more fully described by Bentley & Clack [2004; 2005].

Organisms in nature exhibit complex adaptive behaviour that far surpasses the ability of current state-of-the-art autonomous software and robotics. Our research focuses on morphological adaptation, the continuous lifetime re- configuration of phenotypic form (shape) exhibited by natural systems in or- der to continue to survive in a changing environment. Many unicell organisms exhibit complex adaptations of their shape in rapid response to environmental changes — e.g. fibroblast cells change shape to assist movement during wound healing, and immune system cells change shape to eat invading bacteria — even though they have no centralized control system. We aim to understand the underlying mechanisms and principles that govern this adaptive behaviour, to explore the concept of morphological adaptation as a mapping from envi- ronment to phenotype rather than merely from genotype to phenotype, and to draw inspiration from those mechanisms to improve the adaptive behaviour of artificial systems.

The detailed spatiotemporal aspects of morphogenesis are difficult to com- pute using partial differential equations and so we turned to a bioscience com- puting technique; a cellular automaton and agent-based computing using a very large number of simple agents (‘swarm’ agents).

1.2.1 The artificial cytoskeleton

Our mechanism, the ‘artificial cytoskeleton’, is closely modelled on the eu- karyotic cytoskeleton, a complex, dynamic network of protein filaments which extends throughout the cytoplasm and which gives the cell dynamic structure levels — thus, an understanding of dynamic emergence in complex hierarchies is a fundamental step in understanding the underlying mechanisms of biology.

Christopher D. Clack

(27)

9 In particular actin cytoskeleton microfilaments are involved

3D cellular automaton (CA) rules to allow proteins to exist and interact with their 26 nearest neighbours in a 3D voxellated environment. The agent-based swarm technique permits the modelling and tracking of individual components and their interactions. The CA simplifies visualization, supports 3D spatial placement and movement, and reduces system complexity. The combination of the two techniques (agent-based swarm and CA) provides opportunities for optimizing computational overhead (e.g. it is not always necessary to compute interactions for all cells in the CA — only those that contain or abut an agent).

The CA rules for chemical diffusion and agent interactions can be checked against current understanding of the biology.

toskeleton via a pathway of protein reactions: the transduction pathway (TP).

See Figure 1.3. For efficiency, the artificial cytoskeleton and transduction path- way comprise only a small selection of proteins — just those necessary for a particular experiment. The artificial cytoskeleton’s non-rigid form permits it to disassemble rapidly and re-form in a more advantageous distribution; it con- stantly responds to environmental cues by reorganizing, i.e. altering the cell’s internal topography and the membrane morphology.

turalproteins (actin and a nucleator), which make up the filaments, and several accessoryproteins, which regulate a filament’s behaviour (e.g. inhibiting, ac- tivating, severing, bundling). Environmental signals filter into the cell via the transduction pathway, affecting concentrations of accessory proteins and struc- tural protein behaviour. The cooperative and competitive interactions of these structural and accessory proteins can dramatically alter the cytoskeleton’s fila- mentous structure, affecting the shape and structure of the cell as a whole, and resulting in rich diversity in cell shape [Alberts et al., 1994].

The protein interactions are defined by a set of functions; these functions encapsulate the complete mapping from environmental cues to cell morphol- ogy (which in turn may affect the environment). We call this function set the

‘environment-phenotype map’ (or ‘E-P map’). Different cell behaviours may require different E-P maps. The following explanation of the underlying mech- anism will focus on the E-P map for chemotaxis; see Figures 1.3 and 1.4.

Each voxel in the cellular automaton contains one of the following units:

1. environment which may contain concentrations of a chemoattractant ‘C’.

2. cytoplasm which may contain concentrations of the protein profilin;

and function.

in rapid changes to membrane shape in response to environmental signals [Alberts et al., 1994]. We use agent-based swarm techniques combined with

receptors (sensors) in the membrane relay external signals to the artificial cy- The artificial cytoskeleton resides within a membrane-bound ‘cell’ and

The underlying mechanism. The artificial cytoskeleton consists of struc- Bioscience Computing and Computational Simulation in Biology

(28)

10

Figure 1.3. A generalized environment-phenotyope map. The cytoskeleton is affected by input from the environment (Env) via the transduction pathway (TP) and can affect the shape of the cell, and thereby also the environment.

Figure 1.4. The environment-phenotype map as by Bentley & Clack [2004] abstracted from the biological pathway for fibroblast chemotaxis. The simplified transduction pathway (TP) contains a receptor and two macromolecules PIP2 and WASP, which convey information to the artificial cytoskeleton (ArtCyto).

Christopher D. Clack

(29)

11 3. an agent which may be either:

actin: which may be in the states S-actin (inactive), P1, P2, or F-actin (in a filament) and which has 2 opposing binding sites (‘+’,‘ ’); or a nucleator: the protein complex ‘Arp 2/3’, which may be switched on

or off and has one binding site.

The interactions of these two agents drive the creation, growth and dis- association of actin filaments. The growth of actin filaments forces local membrane shape changes, therefore altering the cell’s overall shape.

4. cell membrane which may contain a receptor and/or the two transduction pathway proteins WASP and PIP2.

The membrane separates the cell from the environment. Initially, no membrane units contain WASP or PIP2 but each has a probability of containing a receptor.

Cell surface receptors are embedded in the membrane and mediate sig- nals from the external environment to the cytoskeleton. Membrane units containing receptors sum the concentration ofCin their adjacent envi- ronment voxels. If the sum exceeds a threshold, a cascade reaction inside the cell is triggered; WASP and PIP2 are activated for the receptor and for its adjacent membrane voxels. If the receptor deactivates, WASP and PIP2deactivate. See Figure 1.5.

TheWASPproteins, when activated by a receptor, recruit agentsnucle- atorandP1 actinto the membrane (see below for a further explanation of recruitment). A recruited nucleator agent will switch on and recruited P1 actin changes state to P2 actin. Activated PIP2 releases a one-off plume of protein profilin which diffuses through cytoplasm units. Deac- tivated PIP2 causesremovalof all profilin in the membrane unit’s adja- cent cytoplasm voxels [Holt & Koffer, 2001].

Protein behaviour is governed by both general rules and specific rules of interaction. The general rules are:

1. Diffusion:accessory proteins are represented as concentration gradients which diffuse through cytoplasm voxels. Diffusion is calculated as by Glazier & Graner [1993]; each cytoplasm voxel has a protein threshold, the excess being evenly distributed to its cytoplasm neighbours.

2. Random movement:when not bound or stuck, an agent moves randomly.

When it moves to a new position, the protein concentration currently in that position is diffused away and the voxel acquires the agent’s identi- fier; the agent’s previous voxel becomes cytoplasm.

Bioscience Computing and Computational Simulation in Biology

(30)

12

Figure 1.5. Artificial cytoskeleton interactions. Receptors detect chemoattractant, WASP and PIP2activate and cause the cytoskeletal behaviours shown in stages 1 6, see text for details.

3. Recruitment: the biological concept of recruitment of proteins, to a spe- cific proteinS, is modelled as follows: an agent follows random move- ment until it encounters anSin its nearest neighbours. It then can only move such that anSis still in its nearest neighbours. Recruitment stops if there is noSnearest neighbour.

The specific rules of interaction for the chemotaxis environment-phenotype map consist of rules governing actin filament formation (and destruction) and rules governing modifications to the shape of the cell membrane. These are illustrated in Figure 1.5 and the stages are described in detail below:

An actin filament (AF) is created when a nucleator agent combines with an actin agent. Figure 1.5 illustrates a chain of F-actin agents ‘F’ and a nucleator (‘Arp2/3’). Each F-actin agent has two binding sites (‘+’/‘ ’): filament growth occurs at the end with the exposed ‘+’ binding site. Subsequently other actin agents may join the filament by attaching to an actin agent already in the fil- ament. Over time the nucleator disassociates (and un-sticks) from its AF and deactivates (stage 1). Similarly actin in a filament (F-actin) loses affinity for the filament allowing cofilin (a severing protein) to disassociate it; it then gets sequestered and changes to the inactive S-actin state (stage 2). Disassociation always occurs at the filament’s ‘ ’ end. The actin or nucleator agent disassoci- ates with a probability that increases with time spent in a filament. As the ‘+’

end of the filament grows, the ‘ ’ end shrinks and the filament, as a higher level entity, moves towards the membrane.

Actin agents are initiated in the inactive state S-actin; S-actin units sum the concentration of profilin in their nearest neighbours — if it exceeds the threshold then the actin binds to profilin and changes to state P1, removing an amount of profilin from the surrounding cytoplasm (stage 3). P1 actin is recruited to active WASP to form P2 (stage 4). After recruited movement,

– –

Christopher D. Clack

(31)

13 if P2 actin has an actin filament ‘+’ site in its nearest neighbours, it binds to it, changes state to F-actin, releases profilin to the surrounding cytoplasm, and moves to the nearest neighbour cytoplasm voxel that permits its ‘ ’ site to directly abut the actin filament ‘+’ site (stage 5).

A nucleator agent activates when recruited by WASP and then can nucleate (start) actin filaments and set their orientation by binding to a P2 actin agent in its nearest neighbours (also see push-out rule below). If there is a fully bound F-actin nearest neighbour, then a nucleator can also ‘stick’ to it and nucleate a branchactin filament (stage 6 [Alberts et al., 1994]).

There are three interactions affecting the cell membrane:

1. A gap must exist or be created between the AF’s ‘+’ end and the mem- brane to allow P2 actin to bind either to F-actin or a nucleator. Adjacent membrane is ‘pushed-out’ — the membrane voxels become cytoplasm and the adjacent environment voxels become membrane (C is diffused away first).2 The precise biology for this process is unclear [Condeelis, 2001].

2. To keep cytoplasm volume constant following ‘push-out’, the cytoplasm or agent (but not F-actin) voxel within the cell that is furthest from the newly created cytoplasm is replaced with a membrane voxel (any af- fected profilin or agent is displaced).

3. If a membrane unit has no contact with inner cellular units, it is removed (becomes an environment unit); this ensures there are no doubled-up layers of membrane.

The combination of the above three interactions contracts the cell at the opposite side to a leading edge and allows the cell’s centre of mass to move.

Experiments. Chemotaxis experiment.The artificial cytoskeleton was tested in a simple experiment based on animal cell chemotaxis, requiring the cell to undergo transformations in form in response to an external chemical stimulus.

The specific E-P map for our chemotaxis experiment [Bentley & Clack, 2004]

is given in Figure 1.4. The artificial cytoskeleton’s response to the stimulus mimicked that of a real fibroblast cell (Figure 1.6), forming a leading edge with protrusions. It moved towards the chemical source purely by lifetime adapation of shape: see Figure 1.7.

Phagocytosis experiment. In nature, a single adaptive mechanism is able to provide different morphologies in response to different environmental stimuli.

2After implementing this rule, a nucleator would switch off as it would no longer have WASP nearest neighbours, so we permit a nucleator to remain switched on if any of its 26 nearest neighbours or any of their surrounding 98 voxels contain WASP.

Bioscience Computing and Computational Simulation in Biology

(32)

14

For example, compare chemotaxis (movement morphology) with phagocyto- sis (ingestion morphology): these two examples are distinct both topologically and functionally, yet are known to be controlled by the same underlying bi- ological mechanism. In chemotaxis, a cell detects a chemical gradient and transforms its morphology in order to follow it to the source. By contrast, phagocytosis is the process of engulfment of a foreign particle for degradation or ingestion [Castellano et al., 2001]; a fairly universal cell function relying on profound rearrangements of the cell membrane.

In phagocytosis, cell surface receptors trigger and bind to the particle, teth- ering it; this causes reactions involving the same proteins downstream as in chemotaxis, but leading to a different morphology — in this case, internal structure change near the edge touching the particle causes an enclosing con- cave morphology called a ‘phagocytic cup’; see Figure 1.8. The simulated morphology is shown in Figure 1.9 — by comparing the medial axis functions for chemotaxis and phagocytosis shapes, we were able to demonstrate a bifur- cation of morphology based on a single E-P map; this is a clear demonstration of the potential of E-P maps, and of our ability to reproduce the multifunction- ality of nature in our artificial simulations.

Figure 1.6. Figure 1.7.Simulation of chemotaxis

leading edge morphology.

Figure 1.8.

the cup forms around the particle.

Figure 1.9.Simulation of phagocytic cup morphology (particle not shown).

Leading edge morphology (top left) during chemotaxis movement.

Phagocytic cup morphology

Christopher D. Clack

(33)

15

1.3 Impact and future directions for bioscience computing

In the computer sciences. An improved understanding of the internal mech- anisms and organisational principles of adaptive behaviour and lifetime plas- ticity, especially adaptation of morphology, will provide foundational results that are applicable to many forms of adaptive response. The improved under- standing of plasticity will provide the basis for a new breed of software sys- tems that are adaptable to continuously changing, dynamic and unpredictable environments, are robust in the face of unexpected change, and are efficient.

This includes improvements to synthetic systems such as autonomous soft- ware agents (e.g. as used for trading and fund optimisation in the financial markets, which need to adapt to a constantly changing environment — note that ‘shape’ may be remapped into an analogous concept such as asset alloca- tion), automatic design systems for physical artefacts, some automatic systems in clinical medicine, some control systems in the automotive and aeronautical industries, and embodied robots (which currently either have a fixed shape or comprise several modular units that may be dynamically reconfigured).

As in nature, when a software agent alters its ‘shape’, it alters its exposure to and interaction with its environment. Further benefit will be in the area of mechanisms for the distributed control of adaptive response. Overall, the impact of our work will be better designed, more robust, more reliable, more adaptable, more efficient and more effective systems.

In the life sciences.

simulation techniques from our collaborators at the Natural History Museum (NHM), at the UCL Department of Oncology, and elsewhere. For example, with the UCL Department of Oncology we are currently using agent-based swarm techniques to simulate the transport of antibody-based drugs through the extra-cellular matrix during cancer therapy, aiming to improve understand- ing and increase the efficacy of therapy. A particular attraction is the use of computational simulation as a biologist’s ‘hypothesis prototyping tool’, and the fact that the simulation permits the fate of individal molecules to be tracked.

Collaboration with the Natural History Museum (NHM).Our initial work with the NHM was a study of the morphogenesis of diatoms (single-celled algae), whose patterned cell walls are thought to be an adaptive response to their environment. Diatoms are one of the most important groups of primary producers on the planet, which have thousands of forms and behaviours, each adapted to a different environment. If their adaptive response to environmental pressure were better understood, they would be a good bio-indicator of changes occurring in the natural environment [Davey & Crawford, 1986].

The observable diatom cell wall morphologies are not explicable by the electron microscopy studies reveal that the There is much current interest in our computational

physics of diffusion alone;

Bioscience Computing and Computational Simulation in Biology

(34)

16

also incorporate the use of cytoplasmic organelles as moulds for different cell wall components. Our simulation of cell wall morphogenesis used the artifi- cial cytoskeleton to represent the physical position of cell wall and cytoskeletal components, and a genetic algorithm to evolve the cytoskeletal control mech- anism. Our model generated representations of diatom cell walls that were, at each stage of development, consistent with empirical observations and ex- hibited some of the functions of diatom cell walls. More importantly, under-

effects and developmental genetic encoding. This was a significant advance in understanding for the NHM and a fruitful collaboration for both parties, for example leading to three research publications: [Bentley et al., 2005], [Bentley

& Clack, 2004] and [Bentley & Clack, 2005].

We are currently seeking funding to conduct a further experiment to aid the NHM in the understanding of diatom colony behaviour. Certain species of diatom have developed a complex set of interactions during morphogenesis, which allows them to form and disband colonies, triggered by environmental cues and giving them a greater chance of survival (e.g. by altering sinking rates to optimize nutrient and light exposure). In general, a colony-forming diatom will, upon cell division, grow two new cells such that their abutting cell walls interlock and hold the cells together; this continues until the filament reaches a certain average length, then (statistically) the most central dividing cell will divide into two new cells that do not interlock, thus dividing the filament in two [Davey & Crawford, 1986]. Diatom colony formation is an explicit and interesting example of morphological adaptation to environmental changes;

it is a type of cyclomorphosis (where adaptation cycles through two or more forms). There has been a large amount of speculation as to how and why certain species of diatom form colonies; it contributes to current understanding within diatom research, and also provides a good model to improve understanding of the hierarchical adaptive systems that underlie morphological plasticity.

1.4 Summary and conclusions

Bioscience computing draws inspiration from biology to solve computer sci- ence challenges and simultaneously uses new bio-inspired adaptive software to simulate biological systems. It is a rapidly emerging field for interdisciplinary

Biologists have a computational, process-oriented understanding of their subject — they think in terms of objects, interactions and processes: processes are more important than the end result; dynamic behaviour is more impor- cytoskeleton is intimately involved in the patterning of the cell wall and may

improved; e.g. the need to consider not only genetics but also environmental standing of the mechanism of the cytoskeleton during morphogenesis was

sciences.

research, with synergistic benefits for both computer science and the life Christopher D. Clack

(35)

17 tant than equilibrium; and behaviour and interactions ofindividualobjects are important. Computational simulation provides in-silico experiments as a pro- totyping technique for hypothesis formulation that more directly maps to the biologists’ understanding of their subject, and can more directly assist their thought processes. The process-oriented approach to simulating biological complexity leads to an increased understanding of dynamic emergence and regulatory interaction and control: this is a fundamental step towards a future theory of biology.

The artificial cytoskeleton has been presented as an example of the computa- tional simulation techniques of bioscience computing, illustrating real benefits accruing to both computer science and the life sciences.

Acknowledgements

research contributions of the following colleagues are nowledged: Katie Bentley (artifical cytoskeleton, PhD student), Cox (diatom morphology, NHM), Dr. Sylvia Nagl (oncology matical biology) and Manish Patel (model integration, PhD cs.ucl.ac.uk/research/bioscience).

References

Alberts, B., D. Bray, J. Lewis, M. Raff, K. Roberts, and J.D. Watson [1994].Molecular Biology of The Cell. Garland Publishing, 3rd edition.

Anderson, A.R.A., and M.A.J. Chaplain [1998]. Continuous and discrete mathematical models of tumour-induced angiogenesis.Bulletin of Mathematical Biology, 60:857–900.

Araujo, R., and D. McElwain [2004]. A history of the study of solid tumour growth: The con- tribution of mathematical modelling.Bulletin of Mathematical Biology, 66:1039–1091.

Bentley, K., and C.D. Clack [2004]. The artificial cytoskeleton for lifetime adaptation of mor- thesis of Living Systems (ALIFE IX), pages 13–16.

Bentley, K., and C.D. Clack [2005]. Morphological plasticity: Environmentally driven mor- phogenesis.Proceedings of the 8th European Conference on Artificial Life (ECAL 2005), Lecture Notes in Artificial Intelligence, 3630:118–127.

Bentley, K., E. Cox, and P. Bentley [2005]. Nature’s batik: A computer evolution model of diatom valve morphogenesis.Nanoscience and Nanotechnology Journal, 5(1):25–34.

Castellano, F., P. Chavrier, and E. Caron [2001]. Actin dynamics during phagocytosis.Seminars in Immunology, 13:347–355.

Condeelis, J. [2001]. How is actin polymerization nucleatedin vivo?TRENDS in Cell Biology, 11(7):288–293.

Davey, M.C., and R.M. Crawford [1986]. Filament formation in the diatom melosira granulata.

Journal of Phycology, 22:144–150.

Gatenby, R.A., and P.K. Maini [2003]. Mathematical oncology: Cancer summed up.Nature, 421:321–324.

The gratefully ack-

Dr. Eileen and mathe- student).

All are members of the UCL BioScience Computing Interest Group (www.

phology.SODANS Workshop proceedings of the 9th Intl. Conf. on the Simulation and Syn- Bioscience Computing and Computational Simulation in Biology

(36)

18

Giavitto, J-L., C. Godin, O. Michel, and P. Prusunkiewicz [2002].Modelling and Simulation of Biological Processes in the Context of Genomics, chapter Computational Models for Inte- grative and Developmental Biology. Hermes.

Glazier, J.A., and F. Graner [1993]. Simulation of the differential adhesion driven rearrangement of biological cells.Physical Review E, 47(3):2128–2154.

Holt, M.R., and A. Koffer [2001]. Cell motility: Proline-rich proteins promote protrusions.

TRENDS in Cell Biology, 11(1):38–46.

Ideker, T., and D. Lauffenburger [2003]. Building with a scaffold: emerging stategies for high- to low-level cellular modelling.Trends in Biotechnology, 21(6):255–262.

Kirkwood, T.B.L., R.J. Boys, C.S. Gillespie, C.J. Proctor, D.P. Shanley, and D.J. Wilkinson [2003]. Towards an e-biology of ageing: Integrating theory and data.Nature Reviews, Mole- cular Cell Biology, 4:243–249.

Noble, D. [2002]. The rise of computational biology.Nature Reviews, Molecular Cell Biology, 3:460–463.

Patel, M. [2004].Internal report. Department of Clinical Oncology, UCL.

Priami C. (ed.) [2003]. Proc. 1st Workshop Computational Methods in Systems Biology.

Springer.

Scalerandi, M., B. Capogrosso Sansone, C. Benati, and C.A. Condat [2002]. Competition effects in the dynamics of tumor cords.Physical Review E, 65(5 Pt 1):051918.

Succi, S., I.V. Karlin, and H. Chen [2002]. Colloquium: Role of the h theorem in lattice boltz- mann hydrodynamics simulations.Reviews of Modern Physics, 74:1203–1220.

Table 1.1. Glossary of biological terms.

term description

actin A protein that links into chains (polymers), forming microfilaments in muscle and other contractile elements in cells.

allele The precise sequence of nucleotides for a specified gene.

cellular secretion The escape of substances from a cell to its environment.

chemotaxis Migration of cells along a concentration gradient of an attractant.

chromosome The self-replicating genetic structure of cells containing the cellular DNA that contains the linear array of genes.

cytoplasm The contents of a cell (but not including the nucleus).

cytoskeleton A system of molecules within eukaryotic cells providing shape, in- ternal spatial organization, and motility, and may assist in communi- cation with other cells.

diatoms Microscopic algae with cell walls made of silicon and of two separat- ing halves.

eukaryote A cell or organism with a membrane-bound nucleus (and other sub- cellular compartments). Includes all organisms except viruses, bacte- ria, and bluegreen algae.

extracellular matrix (ECM)

Any material produced by cells and secreted into the surrounding medium. The properties of the ECM determine the properties of the tissue (e.g. bone versus tendon) and can also affect the behaviour of cells.

fibroblast A cell found in most tissues of the body, involved in wound repair and closure; they migrate towards the wound site via chemotaxis.

Christopher D. Clack

Références

Documents relatifs

the one developed in [2, 3, 1] uses R -filtrations to interpret the arithmetic volume function as the integral of certain level function on the geometric Okounkov body of the

While manifest cluster-mean centering is known to introduce Nickell’s bias, we also showed how some implementations of the latent mean centering approach in the Bayesian framework

Theorem 2 The expected size of the visibility complex of n uniformly distributed unit balls..

The basic idea is to form a rough skeleton of the level set L(t) using any preliminary estimator of f , and to count the number of connected components of the resulting

One common characteristic of the previous literature on shilling attacks on CF-RS is their focus on assessing the global impact of shilling attacks on different CF models by

This section describes the different event logs and trace clustering techniques that are used, and the components of our approach: how the perturbation will be applied; which

Your station score for ten-minute stations represents how many checklist items you did correctly (80 percent of the station score) and how the examiners rated your performance

Now that the attacker has all of the information he needs, he can start sending packets to the server and take over the session. He spoofs the source information and the