100% found this document useful (1 vote)

20 views76 pages

intelligent engineering systems through artificial neural networks volume 17 Smart systems engineering computational intelligence in architecting complex engineering systems 1st Edition Cihan H. Dagli - The full ebook with complete content is ready for download

The document promotes instant access to various ebooks related to intelligent engineering systems, computational intelligence, and artificial neural networks, available for download at ebookgate.com. It includes links to multiple titles edited by Cihan H. Dagli and others, covering topics such as systems engineering, artificial intelligence, and complex systems. Users can download ebooks in different formats like PDF, ePub, and MOBI.

Uploaded by

romeonremza

We take content rights seriously. If you suspect this is your content, claim it here.

Available Formats

Download as PDF, TXT or read online on Scribd

100% found this document useful (1 vote)

20 views76 pages

intelligent engineering systems through artificial neural networks volume 17 Smart systems engineering computational intelligence in architecting complex engineering systems 1st Edition Cihan H. Dagli - The full ebook with complete content is ready for download

Uploaded by

romeonremza

We take content rights seriously. If you suspect this is your content, claim it here.

Available Formats

Download as PDF, TXT or read online on Scribd

You are on page 1/ 76

Instant Ebook Access, One Click Away – Begin at ebookgate.

com

intelligent engineering systems through artificial

neural networks volume 17 Smart systems
engineering computational intelligence in
architecting complex engineering systems 1st
Edition Cihan H. Dagli
https://ebookgate.com/product/intelligent-engineering-
systems-through-artificial-neural-networks-volume-17-smart-
systems-engineering-computational-intelligence-in-
architecting-complex-engineering-systems-1st-edition-cihan-
h-dagli/

OR CLICK BUTTON

DOWLOAD EBOOK

Get Instant Ebook Downloads – Browse at https://ebookgate.com

Click here to visit ebookgate.com and download ebook now
Instant digital products (PDF, ePub, MOBI) available
Download now and explore formats that suit you...

Computational intelligence in architecting engineering

systems 1st Edition C. H. Dagli

https://ebookgate.com/product/computational-intelligence-in-
architecting-engineering-systems-1st-edition-c-h-dagli/

ebookgate.com

Case Studies in System of Systems Enterprise Systems and

Complex Systems Engineering 1st Edition Alex Gorod

https://ebookgate.com/product/case-studies-in-system-of-systems-
enterprise-systems-and-complex-systems-engineering-1st-edition-alex-
gorod/
ebookgate.com

Artificial Intelligence for Intelligent Systems

Fundamentals Challenges and Applications Intelligent Data
Driven Systems and Artificial Intelligence 1st Edition
Inam Ullah Khan
https://ebookgate.com/product/artificial-intelligence-for-intelligent-
systems-fundamentals-challenges-and-applications-intelligent-data-
driven-systems-and-artificial-intelligence-1st-edition-inam-ullah-
khan/
ebookgate.com

Managing and Engineering Complex Technological Systems 1st

Edition Stauber

https://ebookgate.com/product/managing-and-engineering-complex-
technological-systems-1st-edition-stauber/

ebookgate.com
System of Systems Engineering Wiley Series in Systems
Engineering and Management 1st Edition Mohammad Jamshidi

https://ebookgate.com/product/system-of-systems-engineering-wiley-
series-in-systems-engineering-and-management-1st-edition-mohammad-
jamshidi/
ebookgate.com

Agent Directed Simulation and Systems Engineering Wiley

Series in Systems Engineering and Management 1st Edition
Levent Yilmaz
https://ebookgate.com/product/agent-directed-simulation-and-systems-
engineering-wiley-series-in-systems-engineering-and-management-1st-
edition-levent-yilmaz/
ebookgate.com

Fuzzy Engineering Expert Systems with Neural Network

Applications 1st Edition Adedeji Bodunde Badiru

https://ebookgate.com/product/fuzzy-engineering-expert-systems-with-
neural-network-applications-1st-edition-adedeji-bodunde-badiru/

ebookgate.com

Molecular Systems Engineering 1st Edition Efstratios

Pistikopoulos

https://ebookgate.com/product/molecular-systems-engineering-1st-
edition-efstratios-pistikopoulos/

ebookgate.com

Spacecraft Systems Engineering 4th Edition Peter Fortescue

https://ebookgate.com/product/spacecraft-systems-engineering-4th-
edition-peter-fortescue/

ebookgate.com
SMART SYSTEMS ENGINEERING: SMART SYSTEMS ENGINEERING:
COMPUTATIONAL INTELLIGENCE IN ARCHITECTING COMPUTATIONAL INTELLIGENCE IN ARCHITECTING
COMPLEX ENGINEERING SYSTEMS COMPLEX ENGINEERING SYSTEMS

VOLUME 17 VOLUME 17

SMART SYSTEMS ENGINEERING: SMART SYSTEMS ENGINEERING:

COMPUTATIONAL INTELLIGENCE IN ARCHITECTING COMPUTATIONAL INTELLIGENCE IN ARCHITECTING
COMPLEX ENGINEERING SYSTEMS COMPLEX ENGINEERING SYSTEMS

VOLUME 17 VOLUME 17

Downloaded From: http://ebooks.asmedigitalcollection.asme.org/ on 01/28/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

ASME PRESS SERIES ON INTELLIGENT ASME PRESS SERIES ON INTELLIGENT
ENGINEERING SYSTEMS THROUGH ARTIFICIAL ENGINEERING SYSTEMS THROUGH ARTIFICIAL
NEURAL NETWORKS NEURAL NETWORKS

EDITOR EDITOR
C.H. Dagli, Editor, University of Missouri-Rolla, Rolla, Missouri, USA C.H. Dagli, Editor, University of Missouri-Rolla, Rolla, Missouri, USA

Intelligent Engineering Systems Through Artificial Neural Networks, Volume 1, Intelligent Engineering Systems Through Artificial Neural Networks, Volume 1,
edited by Cihan H. Dagli, Soundar R.T. Kumara, and Yung C. Shin, 1991 edited by Cihan H. Dagli, Soundar R.T. Kumara, and Yung C. Shin, 1991

Intelligent Engineering Systems Through Artificial Neural Networks, Volume 2, Intelligent Engineering Systems Through Artificial Neural Networks, Volume 2,
edited by Cihan H. Dagli, Laura I. Burke, and Yung C. Shin, 1992 edited by Cihan H. Dagli, Laura I. Burke, and Yung C. Shin, 1992

Intelligent Engineering Systems Through Artificial Neural Networks, Volume 3, Intelligent Engineering Systems Through Artificial Neural Networks, Volume 3,
edited by Cihan H. Dagli, Laura I. Burke, Benito Fernandez, and Joydeep Ghosh, edited by Cihan H. Dagli, Laura I. Burke, Benito Fernandez, and Joydeep Ghosh,
1993 1993

Intelligent Engineering Systems Through Artificial Neural Networks, Volume 4, Intelligent Engineering Systems Through Artificial Neural Networks, Volume 4,
edited by Cihan H. Dagli, Benito Fernandez, Joydeep Ghosh, and R.T. Soundar edited by Cihan H. Dagli, Benito Fernandez, Joydeep Ghosh, and R.T. Soundar
Kumara, 1994 Kumara, 1994

Intelligent Engineering Systems Through Artificial Neural Networks, Volume 5, Intelligent Engineering Systems Through Artificial Neural Networks, Volume 5,
edited by Cihan H. Dagli, Metin Akay, C.L. Phillip Chen, Benito Fernandez, and edited by Cihan H. Dagli, Metin Akay, C.L. Phillip Chen, Benito Fernandez, and
Joydeep Ghosh, 1995 Joydeep Ghosh, 1995

Intelligent Engineering Systems Through Artificial Neural Networks, Volume 6, Intelligent Engineering Systems Through Artificial Neural Networks, Volume 6,
edited by Cihan H. Dagli, Metin Akay, C.L. Phillip Chen, Benito Fernandez, and edited by Cihan H. Dagli, Metin Akay, C.L. Phillip Chen, Benito Fernandez, and
Joydeep Ghosh, 1996 Joydeep Ghosh, 1996

Intelligent Engineering Systems Through Artificial Neural Networks: Smart Intelligent Engineering Systems Through Artificial Neural Networks: Smart
Engineering Systems: Neural Networks, Fuzzy Logic, Data Mining and Evolutionary Engineering Systems: Neural Networks, Fuzzy Logic, Data Mining and Evolutionary
Programming, Volume 7, edited by Cihan H. Dagli, Metin Akay, Okan Ersoy, Benito Programming, Volume 7, edited by Cihan H. Dagli, Metin Akay, Okan Ersoy, Benito
Fernandez, and Alice Smith, 1997 Fernandez, and Alice Smith, 1997

Intelligent Engineering Systems Through Artificial Neural Networks: Smart Intelligent Engineering Systems Through Artificial Neural Networks: Smart
Engineering Systems: Neural Networks, Fuzzy Logic, Evolutionary Programming, Engineering Systems: Neural Networks, Fuzzy Logic, Evolutionary Programming,
Data Mining and Rough Sets, Volume 8, edited by Cihan H. Dagli, Metin Akay, Data Mining and Rough Sets, Volume 8, edited by Cihan H. Dagli, Metin Akay,
Anna L. Buczak, Okan Ersoy, and Benito Fernandez, 1998 Anna L. Buczak, Okan Ersoy, and Benito Fernandez, 1998

Intelligent Engineering Systems Through Artificial Neural Networks: Smart Intelligent Engineering Systems Through Artificial Neural Networks: Smart
Engineering System Design: Neural Networks, Fuzzy Logic, Evolutionary Engineering System Design: Neural Networks, Fuzzy Logic, Evolutionary
Programming, Data Mining and Complex Systems, Volume 9, edited by Cihan H. Programming, Data Mining and Complex Systems, Volume 9, edited by Cihan H.
Dagli, Anna L. Buczak, Joydeep Ghosh, Mark Embrechts and Okan Ersoy, 1999 Dagli, Anna L. Buczak, Joydeep Ghosh, Mark Embrechts and Okan Ersoy, 1999

Intelligent Engineering Systems Through Artificial Neural Networks: Smart Intelligent Engineering Systems Through Artificial Neural Networks: Smart
Engineering System Design: Neural Networks, Fuzzy Logic, Evolutionary Engineering System Design: Neural Networks, Fuzzy Logic, Evolutionary
Programming, Data Mining and Complex Systems, Volume 10, edited by Cihan H. Programming, Data Mining and Complex Systems, Volume 10, edited by Cihan H.
Dagli, Anna L. Buczak, Joydeep Ghosh, Mark Embrechts, Okan Ersoy and Dagli, Anna L. Buczak, Joydeep Ghosh, Mark Embrechts, Okan Ersoy and
Stephen Kercel, 2000 Stephen Kercel, 2000

ASME PRESS SERIES ON INTELLIGENT ASME PRESS SERIES ON INTELLIGENT

ENGINEERING SYSTEMS THROUGH ARTIFICIAL ENGINEERING SYSTEMS THROUGH ARTIFICIAL
NEURAL NETWORKS NEURAL NETWORKS

EDITOR EDITOR
C.H. Dagli, Editor, University of Missouri-Rolla, Rolla, Missouri, USA C.H. Dagli, Editor, University of Missouri-Rolla, Rolla, Missouri, USA

Intelligent Engineering Systems Through Artificial Neural Networks: Smart Intelligent Engineering Systems Through Artificial Neural Networks: Smart
Engineering Systems: Neural Networks, Fuzzy Logic, Evolutionary Programming, Engineering Systems: Neural Networks, Fuzzy Logic, Evolutionary Programming,
Data Mining and Rough Sets, Volume 8, edited by Cihan H. Dagli, Metin Akay, Data Mining and Rough Sets, Volume 8, edited by Cihan H. Dagli, Metin Akay,
Anna L. Buczak, Okan Ersoy, and Benito Fernandez, 1998 Anna L. Buczak, Okan Ersoy, and Benito Fernandez, 1998

Intelligent Engineering Systems Through Artificial Neural Networks: Smart Intelligent Engineering Systems Through Artificial Neural Networks: Smart
Engineering System Design: Neural Networks, Fuzzy Logic, Evolutionary Engineering System Design: Neural Networks, Fuzzy Logic, Evolutionary
Programming, Data Mining and Complex Systems, Volume 10, edited by Cihan H. Programming, Data Mining and Complex Systems, Volume 10, edited by Cihan H.
Dagli, Anna L. Buczak, Joydeep Ghosh, Mark Embrechts, Okan Ersoy and Dagli, Anna L. Buczak, Joydeep Ghosh, Mark Embrechts, Okan Ersoy and
Stephen Kercel, 2000 Stephen Kercel, 2000

Downloaded From: http://ebooks.asmedigitalcollection.asme.org/ on 01/28/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

Intelligent Engineering Systems Through Artificial Neural Networks: Smart Intelligent Engineering Systems Through Artificial Neural Networks: Smart
Engineering System Design: Neural Networks, Fuzzy Logic, Evolutionary Engineering System Design: Neural Networks, Fuzzy Logic, Evolutionary
Programming, Data Mining and Complex Systems, Volume 11, edited by Cihan H. Programming, Data Mining and Complex Systems, Volume 11, edited by Cihan H.
Dagli, Anna L. Buczak, Joydeep Ghosh, Mark Embrechts, Okan Ersoy and Dagli, Anna L. Buczak, Joydeep Ghosh, Mark Embrechts, Okan Ersoy and
Stephen Kercel, 2001 Stephen Kercel, 2001

Intelligent Engineering Systems Through Artificial Neural Networks: Smart Intelligent Engineering Systems Through Artificial Neural Networks: Smart
Engineering System Design: Neural Networks, Fuzzy Logic, Evolutionary Engineering System Design: Neural Networks, Fuzzy Logic, Evolutionary
Programming, Complex Systems and Artificial Life, Volume 12, edited by Cihan H. Programming, Complex Systems and Artificial Life, Volume 12, edited by Cihan H.
Dagli, Anna L. Buczak, Joydeep Ghosh, Mark Embrechts, Okan Ersoy and Dagli, Anna L. Buczak, Joydeep Ghosh, Mark Embrechts, Okan Ersoy and
Stephen Kercel, 2002 Stephen Kercel, 2002

Intelligent Engineering Systems Through Artificial Neural Networks: Smart Intelligent Engineering Systems Through Artificial Neural Networks: Smart
Engineering System Design: Neural Networks, Fuzzy Logic, Evolutionary Engineering System Design: Neural Networks, Fuzzy Logic, Evolutionary
Programming, Complex Systems, and Artificial Life, Volume 13, edited by Cihan H. Programming, Complex Systems, and Artificial Life, Volume 13, edited by Cihan H.
Dagli, Anna L. Buczak, Joydeep Ghosh, Mark Embrechts, and Okan Ersoy, 2003 Dagli, Anna L. Buczak, Joydeep Ghosh, Mark Embrechts, and Okan Ersoy, 2003

Intelligent Engineering Systems Through Artificial Neural Networks: Smart Intelligent Engineering Systems Through Artificial Neural Networks: Smart
Engineering System Design: Neural Networks, Fuzzy Logic, Evolutionary Engineering System Design: Neural Networks, Fuzzy Logic, Evolutionary
Programming, Complex Systems and Artificial Life, Volume 14, edited by Cihan H. Programming, Complex Systems and Artificial Life, Volume 14, edited by Cihan H.
Dagli, Anna L. Buczak, David L. Enke, Mark Embrechts and Okan Ersoy, 2004 Dagli, Anna L. Buczak, David L. Enke, Mark Embrechts and Okan Ersoy, 2004

Intelligent Engineering Systems Through Artificial Neural Networks: Smart Intelligent Engineering Systems Through Artificial Neural Networks: Smart
Engineering System Design: Neural Networks, Evolutionary Programming, Data Engineering System Design: Neural Networks, Evolutionary Programming, Data
Mining, and Artificial Life, Volume 15, edited by Cihan H. Dagli, Anna L. Buczak, Mining, and Artificial Life, Volume 15, edited by Cihan H. Dagli, Anna L. Buczak,
David L. Enke, Mark Embrechts, and Okan Ersoy, 2005 David L. Enke, Mark Embrechts, and Okan Ersoy, 2005

Intelligent Engineering Systems Through Artificial Neural Networks: Smart Systems Intelligent Engineering Systems Through Artificial Neural Networks: Smart Systems
Engineering: Infra-Structure Systems Engineering, Bio-Informatics and Engineering: Infra-Structure Systems Engineering, Bio-Informatics and
Computational Biology, Evolutionary Computation, Volume 16, edited by Cihan H. Computational Biology, Evolutionary Computation, Volume 16, edited by Cihan H.
Dagli, Anna L. Buczak, David L. Enke, Mark Embrechts, and Okan Ersoy, 2006 Dagli, Anna L. Buczak, David L. Enke, Mark Embrechts, and Okan Ersoy, 2006

Intelligent Engineering Systems Through Artificial Neural Networks: Smart Systems Intelligent Engineering Systems Through Artificial Neural Networks: Smart Systems
Engineering: Computational Intelligence in Architecting Complex Engineering Engineering: Computational Intelligence in Architecting Complex Engineering
Systems, Volume 17, edited by Cihan H. Dagli, Anna L. Buczak, David L. Enke, Systems, Volume 17, edited by Cihan H. Dagli, Anna L. Buczak, David L. Enke,
Mark Embrechts, and Okan Ersoy, 2007 Mark Embrechts, and Okan Ersoy, 2007

Intelligent Engineering Systems Through Artificial Neural Networks: Smart Intelligent Engineering Systems Through Artificial Neural Networks: Smart
Engineering System Design: Neural Networks, Fuzzy Logic, Evolutionary Engineering System Design: Neural Networks, Fuzzy Logic, Evolutionary
Programming, Data Mining and Complex Systems, Volume 11, edited by Cihan H. Programming, Data Mining and Complex Systems, Volume 11, edited by Cihan H.
Dagli, Anna L. Buczak, Joydeep Ghosh, Mark Embrechts, Okan Ersoy and Dagli, Anna L. Buczak, Joydeep Ghosh, Mark Embrechts, Okan Ersoy and
Stephen Kercel, 2001 Stephen Kercel, 2001

Intelligent Engineering Systems Through Artificial Neural Networks: Smart Intelligent Engineering Systems Through Artificial Neural Networks: Smart
Engineering System Design: Neural Networks, Fuzzy Logic, Evolutionary Engineering System Design: Neural Networks, Fuzzy Logic, Evolutionary
Programming, Complex Systems and Artificial Life, Volume 12, edited by Cihan H. Programming, Complex Systems and Artificial Life, Volume 12, edited by Cihan H.
Dagli, Anna L. Buczak, Joydeep Ghosh, Mark Embrechts, Okan Ersoy and Dagli, Anna L. Buczak, Joydeep Ghosh, Mark Embrechts, Okan Ersoy and
Stephen Kercel, 2002 Stephen Kercel, 2002

Intelligent Engineering Systems Through Artificial Neural Networks: Smart Intelligent Engineering Systems Through Artificial Neural Networks: Smart
Engineering System Design: Neural Networks, Fuzzy Logic, Evolutionary Engineering System Design: Neural Networks, Fuzzy Logic, Evolutionary
Programming, Complex Systems, and Artificial Life, Volume 13, edited by Cihan H. Programming, Complex Systems, and Artificial Life, Volume 13, edited by Cihan H.
Dagli, Anna L. Buczak, Joydeep Ghosh, Mark Embrechts, and Okan Ersoy, 2003 Dagli, Anna L. Buczak, Joydeep Ghosh, Mark Embrechts, and Okan Ersoy, 2003

Intelligent Engineering Systems Through Artificial Neural Networks: Smart Intelligent Engineering Systems Through Artificial Neural Networks: Smart
Engineering System Design: Neural Networks, Fuzzy Logic, Evolutionary Engineering System Design: Neural Networks, Fuzzy Logic, Evolutionary
Programming, Complex Systems and Artificial Life, Volume 14, edited by Cihan H. Programming, Complex Systems and Artificial Life, Volume 14, edited by Cihan H.
Dagli, Anna L. Buczak, David L. Enke, Mark Embrechts and Okan Ersoy, 2004 Dagli, Anna L. Buczak, David L. Enke, Mark Embrechts and Okan Ersoy, 2004

Intelligent Engineering Systems Through Artificial Neural Networks: Smart Intelligent Engineering Systems Through Artificial Neural Networks: Smart
Engineering System Design: Neural Networks, Evolutionary Programming, Data Engineering System Design: Neural Networks, Evolutionary Programming, Data
Mining, and Artificial Life, Volume 15, edited by Cihan H. Dagli, Anna L. Buczak, Mining, and Artificial Life, Volume 15, edited by Cihan H. Dagli, Anna L. Buczak,
David L. Enke, Mark Embrechts, and Okan Ersoy, 2005 David L. Enke, Mark Embrechts, and Okan Ersoy, 2005

Intelligent Engineering Systems Through Artificial Neural Networks: Smart Systems Intelligent Engineering Systems Through Artificial Neural Networks: Smart Systems
Engineering: Computational Intelligence in Architecting Complex Engineering Engineering: Computational Intelligence in Architecting Complex Engineering
Systems, Volume 17, edited by Cihan H. Dagli, Anna L. Buczak, David L. Enke, Systems, Volume 17, edited by Cihan H. Dagli, Anna L. Buczak, David L. Enke,
Mark Embrechts, and Okan Ersoy, 2007 Mark Embrechts, and Okan Ersoy, 2007

Downloaded From: http://ebooks.asmedigitalcollection.asme.org/ on 01/28/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

Downloaded From: http://ebooks.asmedigitalcollection.asme.org/ on 01/28/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use
SMART SYSTEMS ENGINEERING: SMART SYSTEMS ENGINEERING:
Computational Intelligence in Architecting Computational Intelligence in Architecting
Complex Engineering Systems Complex Engineering Systems

VOLUME 17 VOLUME 17

Proceedings of the Artificial Neural Networks in Engineering Conference Proceedings of the Artificial Neural Networks in Engineering Conference
(ANNIE 2007) held November 11-14, 2007, in St. Louis, Missouri, U.S.A. (ANNIE 2007) held November 11-14, 2007, in St. Louis, Missouri, U.S.A.

EDITORS EDITORS
Cihan H. Dagli Cihan H. Dagli
University of Missouri-Rolla, Missouri University of Missouri-Rolla, Missouri
Rolla, Missouri Rolla, Missouri

Anna L. Buczak Anna L. Buczak

Sarnoff Corp. Sarnoff Corp.
Princeton, New Jersey Princeton, New Jersey

David L. Enke David L. Enke

University of Tulsa, Oklahoma University of Tulsa, Oklahoma
Tulsa, Oklahoma Tulsa, Oklahoma

Mark Embrechts Mark Embrechts

Rensselaer Polytechnic Institute, RPI Rensselaer Polytechnic Institute, RPI
Troy, New York Troy, New York

Okan Ersoy Okan Ersoy

Purdue University Purdue University
West Lafayette, Indiana West Lafayette, Indiana

ASME PRESS NEW YORK 2007 ASME PRESS NEW YORK 2007

SMART SYSTEMS ENGINEERING: SMART SYSTEMS ENGINEERING:

Computational Intelligence in Architecting Computational Intelligence in Architecting
Complex Engineering Systems Complex Engineering Systems

VOLUME 17 VOLUME 17

EDITORS EDITORS
Cihan H. Dagli Cihan H. Dagli
University of Missouri-Rolla, Missouri University of Missouri-Rolla, Missouri
Rolla, Missouri Rolla, Missouri

Anna L. Buczak Anna L. Buczak

Sarnoff Corp. Sarnoff Corp.
Princeton, New Jersey Princeton, New Jersey

David L. Enke David L. Enke

University of Tulsa, Oklahoma University of Tulsa, Oklahoma
Tulsa, Oklahoma Tulsa, Oklahoma

Mark Embrechts Mark Embrechts

Rensselaer Polytechnic Institute, RPI Rensselaer Polytechnic Institute, RPI
Troy, New York Troy, New York

Okan Ersoy Okan Ersoy

Purdue University Purdue University
West Lafayette, Indiana West Lafayette, Indiana

ASME PRESS NEW YORK 2007 ASME PRESS NEW YORK 2007

Downloaded From: http://ebooks.asmedigitalcollection.asme.org/ on 01/28/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

www.asme.org www.asme.org

All rights reserved. Printed in the United States of America. Except as All rights reserved. Printed in the United States of America. Except as
permitted under the United States Copyright Act of 1976, no part of permitted under the United States Copyright Act of 1976, no part of
this publication may be reproduced or distributed in any form or by this publication may be reproduced or distributed in any form or by
any means, or stored in a database or retrieval system, without the any means, or stored in a database or retrieval system, without the
prior written permission of the publisher. prior written permission of the publisher.

INFORMATION CONTAINED IN THIS WORK HAS BEEN OBTAINED BY INFORMATION CONTAINED IN THIS WORK HAS BEEN OBTAINED BY
THE AMERICAN SOCIETY OF MECHANICAL ENGINEERS FROM THE AMERICAN SOCIETY OF MECHANICAL ENGINEERS FROM
SOURCES BELIEVED TO BE RELIABLE. HOWEVER, NEITHER ASME SOURCES BELIEVED TO BE RELIABLE. HOWEVER, NEITHER ASME
NOR ITS AUTHORS OR EDITORS GUARANTEE THE ACCURACY OR NOR ITS AUTHORS OR EDITORS GUARANTEE THE ACCURACY OR
COMPLETENESS OF ANY INFORMATION PUBLISHED IN THIS WORK. COMPLETENESS OF ANY INFORMATION PUBLISHED IN THIS WORK.
NEITHER ASME NOR ITS AUTHORS AND EDITORS SHALL BE NEITHER ASME NOR ITS AUTHORS AND EDITORS SHALL BE
RESPONSIBLE FOR ANY ERRORS, OMISSIONS, OR DAMAGES RESPONSIBLE FOR ANY ERRORS, OMISSIONS, OR DAMAGES
ARISING OUT OF THE USE OF THIS INFORMATION. THE WORK IS ARISING OUT OF THE USE OF THIS INFORMATION. THE WORK IS
PUBLISHED WITH THE UNDERSTANDING THAT ASME AND ITS PUBLISHED WITH THE UNDERSTANDING THAT ASME AND ITS
AUTHORS AND EDITORS ARE SUPPLYING INFORMATION BUT ARE AUTHORS AND EDITORS ARE SUPPLYING INFORMATION BUT ARE
NOT ATTEMPTING TO RENDER ENGINEERING OR OTHER NOT ATTEMPTING TO RENDER ENGINEERING OR OTHER
PROFESSIONAL SERVICES. IF SUCH ENGINEERING OR PROFESSIONAL SERVICES. IF SUCH ENGINEERING OR
PROFESSIONAL SERVICES ARE REQUIRED, THE ASSISTANCE OF AN PROFESSIONAL SERVICES ARE REQUIRED, THE ASSISTANCE OF AN
APPROPRIATE PROFESSIONAL SHOULD BE SOUGHT. APPROPRIATE PROFESSIONAL SHOULD BE SOUGHT.

ASME shall not be responsible for statements or opinions advanced in ASME shall not be responsible for statements or opinions advanced in
papers or . . . printed in its publications (B7.1.3). Statement from the papers or . . . printed in its publications (B7.1.3). Statement from the
Bylaws. Bylaws.

For authorization to photocopy material for internal or personal use For authorization to photocopy material for internal or personal use
under those circumstances not falling within the fair use provisions of under those circumstances not falling within the fair use provisions of
the Copyright Act, contact the Copyright Clearance Center (CCC), 222 the Copyright Act, contact the Copyright Clearance Center (CCC), 222
Rosewood Drive, Danvers, MA 01923, tel: 978-750-8400, Rosewood Drive, Danvers, MA 01923, tel: 978-750-8400,
www.copyright.com. www.copyright.com.

Requests for special permission or bulk reproduction should be addressed to Requests for special permission or bulk reproduction should be addressed to
the ASME Publishing Department the ASME Publishing Department

Library of Congress 92-30949 Library of Congress 92-30949

ISBN: 0-7918-0265-5 ISBN: 0-7918-0265-5
ASME Order No. 802655 ASME Order No. 802655

www.asme.org www.asme.org

Library of Congress 92-30949 Library of Congress 92-30949

ISBN: 0-7918-0265-5 ISBN: 0-7918-0265-5
ASME Order No. 802655 ASME Order No. 802655

Downloaded From: http://ebooks.asmedigitalcollection.asme.org/ on 01/28/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

vii vii

Downloaded From: http://ebooks.asmedigitalcollection.asme.org/ on 01/28/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

viii viii

Downloaded From: http://ebooks.asmedigitalcollection.asme.org/ on 01/28/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

ix ix

Downloaded From: http://ebooks.asmedigitalcollection.asme.org/ on 01/28/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

x x

Downloaded From: http://ebooks.asmedigitalcollection.asme.org/ on 01/28/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

xi xi

Downloaded From: http://ebooks.asmedigitalcollection.asme.org/ on 01/28/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

xii xii

Downloaded From: http://ebooks.asmedigitalcollection.asme.org/ on 01/28/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

xiii xiii

Downloaded From: http://ebooks.asmedigitalcollection.asme.org/ on 01/28/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

Downloaded From: http://ebooks.asmedigitalcollection.asme.org/ on 01/28/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use
xv xv

xv xv

Downloaded From: http://ebooks.asmedigitalcollection.asme.org/ on 01/28/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

xvi xvi

Downloaded From: http://ebooks.asmedigitalcollection.asme.org/ on 01/28/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

1 1
3 3

MODELING BRAIN ELECTRICAL ACTIVITY INVOLVING MODELING BRAIN ELECTRICAL ACTIVITY INVOLVING
VAGUS NERVE STIMULATION VAGUS NERVE STIMULATION

MARK H. MYERS ROBERT KOZMA MARK H. MYERS ROBERT KOZMA

Computational Neurodynamics Computational Neurodynamics Computational Neurodynamics Computational Neurodynamics
Laboratory Laboratory Laboratory Laboratory
Department of CS, University of Department of CS, University of Department of CS, University of Department of CS, University of
Memphis , Memphis, TN 38152 Memphis, Memphis TN 38152 Memphis , Memphis, TN 38152 Memphis, Memphis TN 38152
mhmyers @memphis.edu [email protected] mhmyers @memphis.edu [email protected]

ABSTRACT ABSTRACT
Experiments involving electroencephalographic (EEG) analysis of Experiments involving electroencephalographic (EEG) analysis of
spatiotemporal structures identified various characteristic features in the spatiotemporal structures identified various characteristic features in the
measured brain activity of human subjects, while awake, at rest, in sleep, and measured brain activity of human subjects, while awake, at rest, in sleep, and
preceding and during epileptic seizure. Gamma power of EEG signals (20Hz to preceding and during epileptic seizure. Gamma power of EEG signals (20Hz to
80 Hz) varies with changes in the cognitive activity and medical conditions of 80 Hz) varies with changes in the cognitive activity and medical conditions of
the subject. Qualitative characterization of the measured effects involves the the subject. Qualitative characterization of the measured effects involves the
Lyapunov analysis in time domain and power spectral analysis in frequency Lyapunov analysis in time domain and power spectral analysis in frequency
domain. Positive Lyapunov exponents and scale free behavior in spectra are domain. Positive Lyapunov exponents and scale free behavior in spectra are
indications of nonlinear chaotic dynamics underlying brains behavior. The indications of nonlinear chaotic dynamics underlying brains behavior. The
present work introduces results obtained by the KIV model, which represents a present work introduces results obtained by the KIV model, which represents a
hierarchical approach to the spatio-temporal dynamic of the limbic system. Our hierarchical approach to the spatio-temporal dynamic of the limbic system. Our
approach models some aspects of the neurosurgical treatment of medically approach models some aspects of the neurosurgical treatment of medically
intractable epilepsy using Vagus Nerve Stimulation (VNS). Several properties intractable epilepsy using Vagus Nerve Stimulation (VNS). Several properties
of brain dynamics during VNS treatment has been reproduced by KIV of brain dynamics during VNS treatment has been reproduced by KIV
simulations, which indicates a potential for future interpretation of VNS simulations, which indicates a potential for future interpretation of VNS
experimental data. experimental data.

INTRODUCTION INTRODUCTION
In this paper, we study the electrical activity of the limbic system, which is a In this paper, we study the electrical activity of the limbic system, which is a
crucial component of the human. In normal healthy state, brains operate in a crucial component of the human. In normal healthy state, brains operate in a
balanced manner and exhibit high-dimensional complex oscillations, which can balanced manner and exhibit high-dimensional complex oscillations, which can
be characterized as chaotic (Freeman, 2000). According to Freeman, the be characterized as chaotic (Freeman, 2000). According to Freeman, the
formation of global amplitude modulation patterns that are integrated into the formation of global amplitude modulation patterns that are integrated into the
entire hemisphere of the brain causes the brain to remain in homogeneous entire hemisphere of the brain causes the brain to remain in homogeneous
neuronal electrical firing states (Freeman, 2006, Freeman, Holmes, West, neuronal electrical firing states (Freeman, 2006, Freeman, Holmes, West,
Vanhatalo, 2005). When the brain is subjected to seizure or stroke, its electrical Vanhatalo, 2005). When the brain is subjected to seizure or stroke, its electrical
activity becomes imbalanced which results in constraints of the original activity becomes imbalanced which results in constraints of the original
highdimensional EEG oscillations. highdimensional EEG oscillations.

We model the balanced, homogenous chaotic background activity in the brain We model the balanced, homogenous chaotic background activity in the brain
through the KIV model, which is a biologically inspired neural network. The through the KIV model, which is a biologically inspired neural network. The
KIV network is comprised of several KIII models linked with distributed and KIV network is comprised of several KIII models linked with distributed and
delayed feedback (Kozma, 2003, Kozma, Freeman, Erdi, 2003, Kozma, Myers, delayed feedback (Kozma, 2003, Kozma, Freeman, Erdi, 2003, Kozma, Myers,

1 1
3 3

MODELING BRAIN ELECTRICAL ACTIVITY INVOLVING MODELING BRAIN ELECTRICAL ACTIVITY INVOLVING
VAGUS NERVE STIMULATION VAGUS NERVE STIMULATION

MARK H. MYERS ROBERT KOZMA MARK H. MYERS ROBERT KOZMA

Downloaded From: http://ebooks.asmedigitalcollection.asme.org/ on 01/28/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

2 2
4 4

2005). We will illustrate the imbalance of electrical brain activity through the 2005). We will illustrate the imbalance of electrical brain activity through the
KIV such as what is found in an EEG during the emergence of a seizure, and KIV such as what is found in an EEG during the emergence of a seizure, and
then reinstate the balanced state of electrical activity through external then reinstate the balanced state of electrical activity through external
stimulation which models VNS treatment. Vagus Nerve Stimulation (VNS) is an stimulation which models VNS treatment. Vagus Nerve Stimulation (VNS) is an
adjunctive treatment for certain types of epilepsy and clinical depression adjunctive treatment for certain types of epilepsy and clinical depression
(Wheless JW, Baumgartner J. 2004). VNS involves sending electric impulses to (Wheless JW, Baumgartner J. 2004). VNS involves sending electric impulses to
the vagus nerve in the neck via a lead implanted under the skin. VNS treatment the vagus nerve in the neck via a lead implanted under the skin. VNS treatment
returns the imbalanced electrical activity of the brain back to a homogeneous returns the imbalanced electrical activity of the brain back to a homogeneous
state (Adamolekun, Jouny, Franaszczuk, 2004). In this work, initial results of state (Adamolekun, Jouny, Franaszczuk, 2004). In this work, initial results of
modeling VNS treatment are introduced. modeling VNS treatment are introduced.

THE KIV MODEL THE KIV MODEL

The KIV model is a biologically inspired neural network (Kozma, Freeman, The KIV model is a biologically inspired neural network (Kozma, Freeman,
Erdi, 2003). K-set family includes hierarchy of K models, including K0, KI, KII, Erdi, 2003). K-set family includes hierarchy of K models, including K0, KI, KII,
KIII, and KIV, see Fig. 1. K-sets represent family of models with increasing KIII, and KIV, see Fig. 1. K-sets represent family of models with increasing
complexity. They represent different aspects of the vertebrate brain. K0 consists complexity. They represent different aspects of the vertebrate brain. K0 consists
of a single node with input and output and a nonlinear transfer function. KI is a of a single node with input and output and a nonlinear transfer function. KI is a
coupling of either exhibitory or inhibitory K0 units. KII is a double layer of both coupling of either exhibitory or inhibitory K0 units. KII is a double layer of both
exhibitory and inhibitory KI units. KIII is three-layer architecture with one or exhibitory and inhibitory KI units. KIII is three-layer architecture with one or
more KIIs connected by feed forward and delayed feed back neural network. more KIIs connected by feed forward and delayed feed back neural network.
KIII model is a working example of the chaotic principles .The KIII model is KIII model is a working example of the chaotic principles .The KIII model is
used for robust pattern recognition. Finally, KIV is the mathematical model of used for robust pattern recognition. Finally, KIV is the mathematical model of
the brain. the brain.

Figure 1 – Illustration of K0, KI, KII, KIII, and KIV sets. Solid and open cycles signify Figure 1 – Illustration of K0, KI, KII, KIII, and KIV sets. Solid and open cycles signify
excitatory and inhibitory populations, respectively. excitatory and inhibitory populations, respectively.

KIV model consists of Hippocampus, Cortical and Amygdala. .It has the KIV model consists of Hippocampus, Cortical and Amygdala. .It has the
functionality of sensory perception and action selection. KIV is a chaotic functionality of sensory perception and action selection. KIV is a chaotic
dynamic memory which encodes sensory information in the form of aperiodic dynamic memory which encodes sensory information in the form of aperiodic
spatial-temporal oscillations of non-linear processing elements. Figure 2 spatial-temporal oscillations of non-linear processing elements. Figure 2
provides a KIV model of the hemisphere, Fig. 2a and b. It has three major parts, provides a KIV model of the hemisphere, Fig. 2a and b. It has three major parts,
cortex, hippocampus and enthorhinal cortex with the amygdala, following the cortex, hippocampus and enthorhinal cortex with the amygdala, following the
limbic system, Fig. 2a. Hippocampus is a KIII neural network which models limbic system, Fig. 2a. Hippocampus is a KIII neural network which models

2 2
4 4

THE KIV MODEL THE KIV MODEL

Downloaded From: http://ebooks.asmedigitalcollection.asme.org/ on 01/28/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

3 3
5 5

proprioception including navigation functions. The cortex is another KIII neural proprioception including navigation functions. The cortex is another KIII neural
network which models sensory processing and pattern recognition in various network which models sensory processing and pattern recognition in various
sensory modalities. The amygdala is a KII neural network which is incorporated sensory modalities. The amygdala is a KII neural network which is incorporated
into the KIV neural network. It is the unit where the activations from the both into the KIV neural network. It is the unit where the activations from the both
KIII are taken and decision is made concerning the next action, based on the KIII are taken and decision is made concerning the next action, based on the
fusion of the signals from the KIII sets. Amygdala is linked with both KIIIs as fusion of the signals from the KIII sets. Amygdala is linked with both KIIIs as
shown in the following figure with some weighted matrix. With proper weight shown in the following figure with some weighted matrix. With proper weight
selection the KIV neural network can maintain some non convergent chaotic selection the KIV neural network can maintain some non convergent chaotic
oscillations among all the components of the system. The activations from oscillations among all the components of the system. The activations from
Hippocampus and Cortical KIII’s are transferred between them through Hippocampus and Cortical KIII’s are transferred between them through
connection weight matrix WA. The activations between Hippocampus and connection weight matrix WA. The activations between Hippocampus and
Amygdala are passed through weighted matrix WB. Finally, the activations Amygdala are passed through weighted matrix WB. Finally, the activations
between Cortex and Amygdala are passed through weight matrix WC. between Cortex and Amygdala are passed through weight matrix WC.

Figure 2a. Schematic illustration of the components of the limbic system. Figure 2a. Schematic illustration of the components of the limbic system.

Figure 2b - The KIV model of the limbic system, consisting of two KIII’s and one KII Figure 2b - The KIV model of the limbic system, consisting of two KIII’s and one KII
interconnected sets. interconnected sets.

The weights between the KIII and KII subcomponents provide the manner in The weights between the KIII and KII subcomponents provide the manner in
which we can model various states of electrical neuronal activity found in a which we can model various states of electrical neuronal activity found in a
human EEG. In this paper, will illustrate how additive inhibitory bias in the human EEG. In this paper, will illustrate how additive inhibitory bias in the

3 3
5 5

Figure 2a. Schematic illustration of the components of the limbic system. Figure 2a. Schematic illustration of the components of the limbic system.

Downloaded From: http://ebooks.asmedigitalcollection.asme.org/ on 01/28/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

4 4
6 6

Amygdala can compensate for simulated imbalance in electrical activity and Amygdala can compensate for simulated imbalance in electrical activity and
reinstate it to a normal and balanced state. reinstate it to a normal and balanced state.

MODELING VAGUS NERVE STIMULATION MODELING VAGUS NERVE STIMULATION

We modify the activity in the KIV model by adjusting the weight connections We modify the activity in the KIV model by adjusting the weight connections
between the Cortex, Hippocampus, and Amygdala. A small perturbation in between the Cortex, Hippocampus, and Amygdala. A small perturbation in
initial conditions can greatly affect the evolution of chaotic systems through initial conditions can greatly affect the evolution of chaotic systems through
time, but once those conditions are estiblished, the future of a chaotic system is time, but once those conditions are estiblished, the future of a chaotic system is
just as deterministic as a non-chaotic system (Kozma, 2003). By introducing a just as deterministic as a non-chaotic system (Kozma, 2003). By introducing a
greater bias to the KII signal we can cause the signal to change its chaotic greater bias to the KII signal we can cause the signal to change its chaotic
nature. nature.
Figures 3 display the basal state A with high dimensional oscillations for a Figures 3 display the basal state A with high dimensional oscillations for a
healthy patient. The Lyapunov exponent is 0.10, showing well-developed chaos. healthy patient. The Lyapunov exponent is 0.10, showing well-developed chaos.
Figure 4 exhibits state B with increase inhibitory connection weights. This state Figure 4 exhibits state B with increase inhibitory connection weights. This state
simulates epileptic conditions. The trajectory becomes less chaotic, and the simulates epileptic conditions. The trajectory becomes less chaotic, and the
Lyapunov exponent is 0.02. Figure 5 shows the case of simulated VNS Lyapunov exponent is 0.02. Figure 5 shows the case of simulated VNS
treatment, when highly chaotic behavior is restored in the model by applying treatment, when highly chaotic behavior is restored in the model by applying
external bias and a dynamic component. This is called state C. The calculated external bias and a dynamic component. This is called state C. The calculated
Lyapunov exponent is 0.12 indicating well-developed chaos. State C is similar Lyapunov exponent is 0.12 indicating well-developed chaos. State C is similar
to conditions that the VNS treatment induces on the EEG activity of the seizure to conditions that the VNS treatment induces on the EEG activity of the seizure
patient. Namely, it forces the system back to an average system state. patient. Namely, it forces the system back to an average system state.
Linear Regression is performed on the PSD of each data channel and the Linear Regression is performed on the PSD of each data channel and the
slope of the PSD has been calculated in log-log coordinates. A relatively linear slope of the PSD has been calculated in log-log coordinates. A relatively linear
log-log behavior indicates scale-free properties in the chaotic time series. The log-log behavior indicates scale-free properties in the chaotic time series. The
slope of the PSD is estimated within the 20-80 Hz range. Our model shows that slope of the PSD is estimated within the 20-80 Hz range. Our model shows that
the slope is -3.59 in experiment A, -4.98 in Experiment B, and -2.90 in the slope is -3.59 in experiment A, -4.98 in Experiment B, and -2.90 in
Experiment C. This behavior is qualitatively agrees with experimental findings Experiment C. This behavior is qualitatively agrees with experimental findings
(Freeman, Holmes, West, Vanhatalo, 2005). (Freeman, Holmes, West, Vanhatalo, 2005).

Figure 3 - Simulations with KIV model in the basal normal state A; upper panel: Figure 3 - Simulations with KIV model in the basal normal state A; upper panel:
activation of a cortical node for a 3 s period; lower panel: return plot in the time- activation of a cortical node for a 3 s period; lower panel: return plot in the time-
delayed phase space. delayed phase space.

4 4
6 6

MODELING VAGUS NERVE STIMULATION MODELING VAGUS NERVE STIMULATION

Downloaded From: http://ebooks.asmedigitalcollection.asme.org/ on 01/28/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

5 5
7 7

Figure 4 - Simulations with KIV model in the simulated seizure state B; notations are Figure 4 - Simulations with KIV model in the simulated seizure state B; notations are
the same as in Figure 3. the same as in Figure 3.

Figure 5 - Simulations with KIV model in the simulated VNS treatment state C. Note Figure 5 - Simulations with KIV model in the simulated VNS treatment state C. Note
that higher complexity chaotic state is restored, as in Fig. 3. that higher complexity chaotic state is restored, as in Fig. 3.

5 5
7 7

Downloaded From: http://ebooks.asmedigitalcollection.asme.org/ on 01/28/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

6 6
8 8

CONCLUSIONS AND FUTURE PERSPECTIVES CONCLUSIONS AND FUTURE PERSPECTIVES

The KIV model can be used to simulate the brain electrical activity as observed The KIV model can be used to simulate the brain electrical activity as observed
in EEG experiments and characterize the various cognitive states, i.e., awake, in EEG experiments and characterize the various cognitive states, i.e., awake,
normal, or seizure. The KIV model is a biologically-inspired, non-linear normal, or seizure. The KIV model is a biologically-inspired, non-linear
dynamic neural network that can model EEG activity through incorporating the dynamic neural network that can model EEG activity through incorporating the
normal and pathological electrical states found in human EEG recordings. normal and pathological electrical states found in human EEG recordings.
Dynamical properties of the simulated data exhibit qualitatively similar behavior Dynamical properties of the simulated data exhibit qualitatively similar behavior
as displayed in EEG recordings on seizure patients with a Vagus Nerve Implant. as displayed in EEG recordings on seizure patients with a Vagus Nerve Implant.
This characterization includes Lyapunov exponent and PSD slope over the This characterization includes Lyapunov exponent and PSD slope over the
gamma band 20Hz – 80 Hz. gamma band 20Hz – 80 Hz.
Our model can aid in developing therapies through the characterization of Our model can aid in developing therapies through the characterization of
various normal and pathological brain states, and potentially predict seizure various normal and pathological brain states, and potentially predict seizure
before its onset. Eventually, this study can provide a customized therapy per before its onset. Eventually, this study can provide a customized therapy per
patient which can be developed as a titration method for dampening a seizure patient which can be developed as a titration method for dampening a seizure
occurrence while preserving/less influencing the overall electrical state of the occurrence while preserving/less influencing the overall electrical state of the
brain. brain.

ACKNOWLEDGEMENTS ACKNOWLEDGEMENTS
Experimental assistance by Hima Puppala, University of Memphis is greatly Experimental assistance by Hima Puppala, University of Memphis is greatly
appreciated. appreciated.

REFERENCES REFERENCES

Jouny CC, Adamolekun B, Franaszczuk PJ, et al. ,2004,, Propagation of partial seizures is associated Jouny CC, Adamolekun B, Franaszczuk PJ, et al. ,2004,, Propagation of partial seizures is associated
with increased signal complexity near seizure focus Epilepsia 45: 61-61 Suppl. 7 with increased signal complexity near seizure focus Epilepsia 45: 61-61 Suppl. 7
Freeman, Walter J., 2000, How Brains Make Up Their Minds. Columbia University Press, 37-60 Freeman, Walter J., 2000, How Brains Make Up Their Minds. Columbia University Press, 37-60
Freeman WJ, 2006, Origin, structure, and role of background EEG activity. Part 4. Neural Frame Freeman WJ, 2006, Origin, structure, and role of background EEG activity. Part 4. Neural Frame
Simulation. Neurophysiol. 117: 572-589. Simulation. Neurophysiol. 117: 572-589.
Freeman WJ, Holmes MD, West GA, Vanhatalo S.2005, Dynamics of human neocortex that Freeman WJ, Holmes MD, West GA, Vanhatalo S.2005, Dynamics of human neocortex that
optimizes its stability and flexibility. Int. J. Intelligent Systems, Vol 21:9 pp 881- 901. optimizes its stability and flexibility. Int. J. Intelligent Systems, Vol 21:9 pp 881- 901.
Kozma R, 2003 Sept, On the constructive role of noise in stabilizing itinerant trajectories in chaotic Kozma R, 2003 Sept, On the constructive role of noise in stabilizing itinerant trajectories in chaotic
dynamical systems. Chaos, Vol. 13:3, pp 1078 – 1089. dynamical systems. Chaos, Vol. 13:3, pp 1078 – 1089.
Kozma R, Freeman WJ, Erdi P, 2003, The KIV Model – Nonlinear Spatio-temporal Dynamics of the Kozma R, Freeman WJ, Erdi P, 2003, The KIV Model – Nonlinear Spatio-temporal Dynamics of the
Primordial Vertebrate Forebrain. Neurocomputing, 52-54, pp. 819-826. Primordial Vertebrate Forebrain. Neurocomputing, 52-54, pp. 819-826.
Kozma R, Myers M, 2005, Analysis of Phase Transitions in KIV with Amygdale during Simulated Kozma R, Myers M, 2005, Analysis of Phase Transitions in KIV with Amygdale during Simulated
Navigational Control. IEEE Inf. Joint Conf. Neural Netw., Montreal, Canada. Navigational Control. IEEE Inf. Joint Conf. Neural Netw., Montreal, Canada.
Wheless JW, Baumgartner J., 2004, Vagus nerve stimulation therapy. Drugs Today (Barc). Wheless JW, Baumgartner J., 2004, Vagus nerve stimulation therapy. Drugs Today (Barc).
40(6):501-15. 40(6):501-15.

6 6
8 8

CONCLUSIONS AND FUTURE PERSPECTIVES CONCLUSIONS AND FUTURE PERSPECTIVES

REFERENCES REFERENCES

Downloaded From: http://ebooks.asmedigitalcollection.asme.org/ on 01/28/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

1 1
9 9

PHASE SYNCHRONIZATION IN MESOSCOPIC PHASE SYNCHRONIZATION IN MESOSCOPIC

ELECTROENCEPHALOGRAM ARRAYS ELECTROENCEPHALOGRAM ARRAYS

JOSE M. RODRIGUEZ ROBERT KOZMA JOSE M. RODRIGUEZ ROBERT KOZMA

Department of Computer Science, Department of Computer Science, Department of Computer Science, Department of Computer Science,
The University of Memphis, The University of Memphis, The University of Memphis, The University of Memphis,
Memphis, TN. Memphis, TN. Memphis, TN. Memphis, TN.

ABSTRACT ABSTRACT
Coordination of neural oscillations generated at different frequencies has been Coordination of neural oscillations generated at different frequencies has been
conjectured to be an important feature of neural activity. In this paper we conjectured to be an important feature of neural activity. In this paper we
investigate the phenomena of phase synchronization in electroencephalogram investigate the phenomena of phase synchronization in electroencephalogram
mesoscopic arrays. The electrical activity of the neurons in the visual cortex of mesoscopic arrays. The electrical activity of the neurons in the visual cortex of
rabbits is recorded and analyzed. An 8*8 array of electrodes is placed on the rabbits is recorded and analyzed. An 8*8 array of electrodes is placed on the
visual cortex of the subject. We consider each electrode as a variable in the visual cortex of the subject. We consider each electrode as a variable in the
phase space of a dynamical system. We decompose each variable in the phase phase space of a dynamical system. We decompose each variable in the phase
space into narrow frequency signals using finite impulse filters according to space into narrow frequency signals using finite impulse filters according to
standard clinical bands. We measure the instantaneous amplitude and phase of standard clinical bands. We measure the instantaneous amplitude and phase of
each electrode signal using the continuous time analytical signal concept. We each electrode signal using the continuous time analytical signal concept. We
conjecture that the dynamics of the visual cortex can be considered a conjecture that the dynamics of the visual cortex can be considered a
metastable system with periods of absolute and relative synchronization. metastable system with periods of absolute and relative synchronization.

INTRODUCTION INTRODUCTION
In this section we introduce the fundamental concepts of dynamical systems In this section we introduce the fundamental concepts of dynamical systems
DS in order to provide the theoretical background for the analysis of phase DS in order to provide the theoretical background for the analysis of phase
synchronization in DS. A DS provides a mathematical description for evolving synchronization in DS. A DS provides a mathematical description for evolving
the state of a system forward in time. The DS can be considered as a flow dx(t) / the state of a system forward in time. The DS can be considered as a flow dx(t) /
dt = F( x(t) ) where x is a n-dimensional vector. Given the initial conditions dt = F( x(t) ) where x is a n-dimensional vector. Given the initial conditions
x(0), we can solve the equations and obtain the future state of the system x(t). x(0), we can solve the equations and obtain the future state of the system x(t).
The n-dimensional state space described by the DS is called the phase space, and The n-dimensional state space described by the DS is called the phase space, and
the path followed by the system from a given initial state in the phase space is the path followed by the system from a given initial state in the phase space is
called trajectory or orbit. The phase space of the DS can be characterized using called trajectory or orbit. The phase space of the DS can be characterized using
the concept of attractors and Fourier analyis. Types of attractors are fixed the concept of attractors and Fourier analyis. Types of attractors are fixed
points, periodic attractors and chaotic or strange attractors (Strogatz 1994). In points, periodic attractors and chaotic or strange attractors (Strogatz 1994). In
Fourier analysis Chutchfield et al. (Crutchfield, Farmer et al. 1980) shows that a Fourier analysis Chutchfield et al. (Crutchfield, Farmer et al. 1980) shows that a
power spectrum with large number of peaks corresponds to attractors with power spectrum with large number of peaks corresponds to attractors with
complicated geometrical structure. A power spectrum with few peaks complicated geometrical structure. A power spectrum with few peaks
corresponds to attractors with less complicated geometrical properties. corresponds to attractors with less complicated geometrical properties.

Biological neural networks can be considered a particular type of DS. We use Biological neural networks can be considered a particular type of DS. We use
electrodes to measure the electrical activity of the brain in the phase space. electrodes to measure the electrical activity of the brain in the phase space.
Each electrode samples the electrical activity of approximately 200 hundreds of Each electrode samples the electrical activity of approximately 200 hundreds of
biological neurons. The phase space obtained in this way has been used in many biological neurons. The phase space obtained in this way has been used in many
ways. Pikovsky (Pikovsky, Rosenblum et al. 2001) use the phase space to ways. Pikovsky (Pikovsky, Rosenblum et al. 2001) use the phase space to
measure the strength of phase synchronization among the objects that compose measure the strength of phase synchronization among the objects that compose
the DS. Freeman (Freeman 2004) uses the phase space in order to analyze the the DS. Freeman (Freeman 2004) uses the phase space in order to analyze the

1 1
9 9

PHASE SYNCHRONIZATION IN MESOSCOPIC PHASE SYNCHRONIZATION IN MESOSCOPIC

ELECTROENCEPHALOGRAM ARRAYS ELECTROENCEPHALOGRAM ARRAYS

JOSE M. RODRIGUEZ ROBERT KOZMA JOSE M. RODRIGUEZ ROBERT KOZMA

Downloaded From: http://ebooks.asmedigitalcollection.asme.org/ on 01/28/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

2 2
10 10

mesoscopic activity of biological neural networks. Kelso (Kelso 1995) uses the mesoscopic activity of biological neural networks. Kelso (Kelso 1995) uses the
concept of metastability to characterize the coordination dynamics of the brain. concept of metastability to characterize the coordination dynamics of the brain.
Metastability is the realization of two opposing tendencies. The tendencies in Metastability is the realization of two opposing tendencies. The tendencies in
the DS components to couple together (phase trapping) and the tendency of the the DS components to couple together (phase trapping) and the tendency of the
components to perform their own innate behavior (phase scattering). components to perform their own innate behavior (phase scattering).

In this paper, we study phase synchronization in mesoscopic arrays. In the first In this paper, we study phase synchronization in mesoscopic arrays. In the first
section we introduce the fundamental concepts related to phase synchronization. section we introduce the fundamental concepts related to phase synchronization.
Hilbert transform and Fourier filtering allow us to characterize the dynamics of Hilbert transform and Fourier filtering allow us to characterize the dynamics of
the neurons. Next we apply these techniques to the study the electrical activity the neurons. Next we apply these techniques to the study the electrical activity
of biological neurons in the visual cortex. We use metastability concepts to of biological neurons in the visual cortex. We use metastability concepts to
characterize the phase transitions. We use the term signal or channel meaning a characterize the phase transitions. We use the term signal or channel meaning a
variable in the DS phase space. variable in the DS phase space.

PHASE SYNCHRONIZATION IN DYNACAL SYSTEMS PHASE SYNCHRONIZATION IN DYNACAL SYSTEMS

To measure the phase coupling between two variables in a DS, we have To measure the phase coupling between two variables in a DS, we have
(Pikovsky, Rosenblum et al. 2001) (Pikovsky, Rosenblum et al. 2001)
\ n ,m (t ) ( nI1 (t ) mI2 (t )) mod( 2S ) (1) \ n ,m (t ) ( nI1 (t ) mI2 (t )) mod( 2S ) (1)
Here n and m are integers. I1 (t ) and I2 (t ) are the phases of each DS variable. Here n and m are integers. I1 (t ) and I2 (t ) are the phases of each DS variable.
For uncoupled variables the distribution of \ n,m is uniform, whereas the For uncoupled variables the distribution of \ n,m is uniform, whereas the
interaction between the variables makes the distribution unimodal. To measure interaction between the variables makes the distribution unimodal. To measure
the phase synchronization strength between the two variables in the DS, we use the phase synchronization strength between the two variables in the DS, we use
the phase coupling index based on the Shannon entropy of the phase distribution the phase coupling index based on the Shannon entropy of the phase distribution
differences (Pikovsky, Rosenblum et al. 1999). Having computed pk in the differences (Pikovsky, Rosenblum et al. 1999). Having computed pk in the
histogram of ȥn,m the Shannon entropy index ȡn,m is computed as histogram of ȥn,m the Shannon entropy index ȡn,m is computed as
N N
U n ,m (ln N ¦ k 1 p k ln p k ) / ln N (2) U n ,m (ln N ¦ k 1 p k ln p k ) / ln N (2)
where N is the number of bins in the histogram and pk is the relative frequency where N is the number of bins in the histogram and pk is the relative frequency
of finding ȥn,m within the k-th bin. of finding ȥn,m within the k-th bin.

The first step in order to analyze the relationships between phases in the DS is to The first step in order to analyze the relationships between phases in the DS is to
estimate the phases. We consider the method based in the construction of the estimate the phases. We consider the method based in the construction of the
complex analytical signal AS z(t) (Boashash 1992; Marple and Lawrance 1999) complex analytical signal AS z(t) (Boashash 1992; Marple and Lawrance 1999)
using the Hilbert transform HT. To implement this method, one has to construct using the Hilbert transform HT. To implement this method, one has to construct
from a scalar signal s(t), a complex signal z(t) from a scalar signal s(t), a complex signal z(t)
z ( t ) s ( t ) iH ( s ( t )) A ( t ) e iI ( t ) (3) z ( t ) s ( t ) iH ( s ( t )) A ( t ) e iI ( t ) (3)
where H(s(t)) is the HT of s(t), A(t) is the analytic amplitude, and I (t ) is the where H(s(t)) is the HT of s(t), A(t) is the analytic amplitude, and I (t ) is the
analytic phase. The HT of s(t) is defined as analytic phase. The HT of s(t) is defined as
f f
H ( s ( t )) (1 / S ) P ³ (s (W ) /( t W )) d W (4) H ( s ( t )) (1 / S ) P ³ (s (W ) /( t W )) d W (4)
f f
where P means that integral is taken in the sense of the Cauchy Principal where P means that integral is taken in the sense of the Cauchy Principal
Value. Practical hints in order to compute and the usage of the HT can be found Value. Practical hints in order to compute and the usage of the HT can be found
in (Boashash 1992; Rosenblum, Pikovsky et al. 2001). Using the straightened in (Boashash 1992; Rosenblum, Pikovsky et al. 2001). Using the straightened

2 2
10 10

PHASE SYNCHRONIZATION IN DYNACAL SYSTEMS PHASE SYNCHRONIZATION IN DYNACAL SYSTEMS

Downloaded From: http://ebooks.asmedigitalcollection.asme.org/ on 01/28/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

3 3
11 11

phases of I (t ) , we can compute the analytic frequency for each variable in the phases of I (t ) , we can compute the analytic frequency for each variable in the
DS phase space as follows DS phase space as follows
d I (t ) df ( t ) (5) d I (t ) df ( t ) (5)
2S 2S
dt dt dt dt

RESULTS – THE THETA BAND (3-7 HZ) RESULTS – THE THETA BAND (3-7 HZ)
The electrical activity of neurons on the surface of the primary visual cortex The electrical activity of neurons on the surface of the primary visual cortex
in rabbits was measured using an 8*8 array of electrodes. Details of the in rabbits was measured using an 8*8 array of electrodes. Details of the
experiments are given by (Freeman 2004); here we summarize the main experiments are given by (Freeman 2004); here we summarize the main
conditions. The space between the electrodes is 0.79 mm covering an entire area conditions. The space between the electrodes is 0.79 mm covering an entire area
of 5.6mm*5.6mm. Data were collected during 6 seconds using a 500 Hz of 5.6mm*5.6mm. Data were collected during 6 seconds using a 500 Hz
sampling frequency. The data were lowpass filtered at 100 Hz. We use finite sampling frequency. The data were lowpass filtered at 100 Hz. We use finite
impulse filters FIR (Porat 1997) to decompose the DS phase space. We use the impulse filters FIR (Porat 1997) to decompose the DS phase space. We use the
standard frequencies defined by the brain clinical community (Freeman 2006). standard frequencies defined by the brain clinical community (Freeman 2006).
Particularly we use the least-square linear phase FIR filters. Choosing correctly Particularly we use the least-square linear phase FIR filters. Choosing correctly
the order of the filter N, these filters have the property of reduce the energy the order of the filter N, these filters have the property of reduce the energy
around a peak. We concentrate our studies in the theta band. around a peak. We concentrate our studies in the theta band.

Fig 1. Computing the AS via the HT (a) Signal s(t). (b) Analytical signal z(t). Fig 1. Computing the AS via the HT (a) Signal s(t). (b) Analytical signal z(t).
(c) Analytical phase ĳ(t) and (d) phase space for z(t). (c) Analytical phase ĳ(t) and (d) phase space for z(t).

We decompose each electrical signal in the array of electrodes using FIR filters. We decompose each electrical signal in the array of electrodes using FIR filters.
The peak frequency is determined as the frequency having the largest spectral The peak frequency is determined as the frequency having the largest spectral
value in the range 3-7 Hz. Theta oscillations are extracted by using a finite value in the range 3-7 Hz. Theta oscillations are extracted by using a finite
impulse filter FIR bandpass filter with cutoff frequencies of ĭpeak ± 2 Hz where impulse filter FIR bandpass filter with cutoff frequencies of ĭpeak ± 2 Hz where
ĭpeak represent a peak. The order of the filter N is set to 181 and the attenuation ĭpeak represent a peak. The order of the filter N is set to 181 and the attenuation
in the passband and stopband are 1 dB and 40 dB respectively. We use least- in the passband and stopband are 1 dB and 40 dB respectively. We use least-
square FIR filters in the decomposition process (Porat 1997). The firls Matlab square FIR filters in the decomposition process (Porat 1997). The firls Matlab
function implement the least-square FIR filters (Losada 2004). Next the AS is function implement the least-square FIR filters (Losada 2004). Next the AS is
calculated for each filtered signal. Figure 1 shows the result. Plots a,b,c and d calculated for each filtered signal. Figure 1 shows the result. Plots a,b,c and d
show one channel in the time domain, the real and the imaginary part of the show one channel in the time domain, the real and the imaginary part of the
analytical signal, the wrapped phases and the real part of z(t) versus the analytical signal, the wrapped phases and the real part of z(t) versus the
imaginary part of z(t). To reduce boundary effects, one half of a second of data imaginary part of z(t). To reduce boundary effects, one half of a second of data
is eliminated at the beginning and ending of each signal. The signal in plot (a) is eliminated at the beginning and ending of each signal. The signal in plot (a)
resembles a sine-like oscillating signal with well defined frequency. Observe resembles a sine-like oscillating signal with well defined frequency. Observe

3 3
11 11

Downloaded From: http://ebooks.asmedigitalcollection.asme.org/ on 01/28/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

4 4
12 12

that the phase has a well defined behavior as shown in plot (c). Plot d shows that the phase has a well defined behavior as shown in plot (c). Plot d shows
that all the phase space of z(t) is well defined. that all the phase space of z(t) is well defined.

Fig 2. Unwrapped phases for all the 64 unwrapped phases. Observe periods of Fig 2. Unwrapped phases for all the 64 unwrapped phases. Observe periods of
synchronization and desynchronization. synchronization and desynchronization.

UNWRAPING THE PHASES UNWRAPING THE PHASES

The analytic phase is computed for all the 64 channels using the AS concept. In The analytic phase is computed for all the 64 channels using the AS concept. In
order to track the behavior of the phase over time, the discontinuous phases are order to track the behavior of the phase over time, the discontinuous phases are
straightened by adding 2ʌ to the arctangent2 function to get the unwrapped straightened by adding 2ʌ to the arctangent2 function to get the unwrapped
analytical phase. The slope of the unwrapped phases gives the average analytical phase. The slope of the unwrapped phases gives the average
frequency over time. The upward and downward deviations show variations in frequency over time. The upward and downward deviations show variations in
the frequency content in theta oscillations. Figure 2 shows the 64 unwrapped the frequency content in theta oscillations. Figure 2 shows the 64 unwrapped
phases. It is interesting to point out in this plot that phase deviations happen in phases. It is interesting to point out in this plot that phase deviations happen in
the array at particular times (See for example times -1.3 seconds and -0.1 the array at particular times (See for example times -1.3 seconds and -0.1
seconds). Here we classify the phase segments in two categories: segments seconds). Here we classify the phase segments in two categories: segments
without phase scatering (absolute phase synchroniation) and segments with without phase scatering (absolute phase synchroniation) and segments with
phase scatering (relative phase synchronization). See Kelso 1995. phase scatering (relative phase synchronization). See Kelso 1995.

SEGMENTS OF ABSOLUTE PHASE SYNCHRONIZATION SEGMENTS OF ABSOLUTE PHASE SYNCHRONIZATION

Figure 3 shows a segment with absolute phase synchronization. The plot in Figure 3 shows a segment with absolute phase synchronization. The plot in
the top-left shows the 64 unwrapped phases. The unwrapped phases look like the top-left shows the 64 unwrapped phases. The unwrapped phases look like
straight lines, resembling sine-like waves with well defined peak in the straight lines, resembling sine-like waves with well defined peak in the
frequency domain. In the top-rigth the unwrapped phase differences are shown. frequency domain. In the top-rigth the unwrapped phase differences are shown.
Here, the phase of channel j (j = 1,..,5,6) has been subtracted from channels k = Here, the phase of channel j (j = 1,..,5,6) has been subtracted from channels k =
j + lag, with lag = 8. Given this difference, we calculate and plot the residues j + lag, with lag = 8. Given this difference, we calculate and plot the residues
after 2ʌ division. The distribution of the phase differences has a peak in zero, after 2ʌ division. The distribution of the phase differences has a peak in zero,
showing statistical phase synchronization. The Shannon synchronization index showing statistical phase synchronization. The Shannon synchronization index
is 0.90. In the bottom-left the analytic frequency is computed for all the is 0.90. In the bottom-left the analytic frequency is computed for all the
channels. The bottom-right shows the distribution of the analytic frequencies. channels. The bottom-right shows the distribution of the analytic frequencies.
The frequencies are smooth out using a moving average filter size 11 samples The frequencies are smooth out using a moving average filter size 11 samples
(44 milliseconds). We lose resolution in time but the analytic frequencies (44 milliseconds). We lose resolution in time but the analytic frequencies
exhibit more regular behavior. exhibit more regular behavior.

4 4
12 12

UNWRAPING THE PHASES UNWRAPING THE PHASES

SEGMENTS OF ABSOLUTE PHASE SYNCHRONIZATION SEGMENTS OF ABSOLUTE PHASE SYNCHRONIZATION

Downloaded From: http://ebooks.asmedigitalcollection.asme.org/ on 01/28/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

5 5
13 13

Fig 3. Absolute synchronization segment. Top-left the unwrapped phases. Fig 3. Absolute synchronization segment. Top-left the unwrapped phases.
Top-right the distribution of phase differences. Bottom-left the frequencies and Top-right the distribution of phase differences. Bottom-left the frequencies and
bottom-right the distribution of frequencies. bottom-right the distribution of frequencies.

SEGMENTS OF RELATIVE PHASE SYNCHRONIZATION SEGMENTS OF RELATIVE PHASE SYNCHRONIZATION

Figure 4 shows a segment with relative phase synchronization. The Figure 4 shows a segment with relative phase synchronization. The
desynchronization periods appear from -1.3 to -1.1 seconds approximately. The desynchronization periods appear from -1.3 to -1.1 seconds approximately. The
plots in the top show the unwrapped phases (left) and the distribution of the plots in the top show the unwrapped phases (left) and the distribution of the
phase’s differences (right). The desynchronization periods last for about 250 phase’s differences (right). The desynchronization periods last for about 250
milliseconds, about one period of oscillation. The Shannon synchronization milliseconds, about one period of oscillation. The Shannon synchronization
index is 0.75. In the bottom of Figure 4 the analytic frequency (left) and the index is 0.75. In the bottom of Figure 4 the analytic frequency (left) and the
distribution of frequencies (right) is shown. Notice that the analytic frequency distribution of frequencies (right) is shown. Notice that the analytic frequency
plot shows negative values. Thus, frequencies f[t] 0 is allowed, while f[t] < 0 plot shows negative values. Thus, frequencies f[t] 0 is allowed, while f[t] < 0
is due to the high sensitivity of the HT (Freeman 2006). is due to the high sensitivity of the HT (Freeman 2006).

Fig 4. A relative desynchronization segment. Top-left the unwrapped phases. Fig 4. A relative desynchronization segment. Top-left the unwrapped phases.
Top-right the distribution of the phase’s differences. Bottom-left the analytic Top-right the distribution of the phase’s differences. Bottom-left the analytic
frequencies. Bottom-right the distribution of the analytic frequencies. frequencies. Bottom-right the distribution of the analytic frequencies.

5 5
13 13

SEGMENTS OF RELATIVE PHASE SYNCHRONIZATION SEGMENTS OF RELATIVE PHASE SYNCHRONIZATION

Downloaded From: http://ebooks.asmedigitalcollection.asme.org/ on 01/28/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

6 6
14 14

DISCUSSION DISCUSSION
Two types of phase synchronization are present in the theta band according Two types of phase synchronization are present in the theta band according
with our findings: Absolute phase synchronization and relative phase with our findings: Absolute phase synchronization and relative phase
synchronization. In absolute phase synchronization periods, the distribution of synchronization. In absolute phase synchronization periods, the distribution of
the phase’s differences has a well defined peak and high Shannon the phase’s differences has a well defined peak and high Shannon
synchronization index. The distribution of phases differences and the synchronization index. The distribution of phases differences and the
distribution of frequencies are narrow. The distribution of phases differences distribution of frequencies are narrow. The distribution of phases differences
does not cover the entire phase interval -2ʌ to 2ʌ. These metastable dynamics does not cover the entire phase interval -2ʌ to 2ʌ. These metastable dynamics
expresses the tendency of the neurons in the theta band to couple together. In expresses the tendency of the neurons in the theta band to couple together. In
relative phase synchronization the distribution of the phase’s differences has a relative phase synchronization the distribution of the phase’s differences has a
well defined peak. The distribution of differences in phase cover the entire well defined peak. The distribution of differences in phase cover the entire
interval of phase differences -2ʌ to 2ʌ. These dynamics correspond to a lower interval of phase differences -2ʌ to 2ʌ. These dynamics correspond to a lower
Shannon synchronization index and broad distribution of phases and Shannon synchronization index and broad distribution of phases and
frequencies. The results provide support for the hypothesis that neurons at the frequencies. The results provide support for the hypothesis that neurons at the
mesoscopic level organize their dynamics by alternating periods of absolute and mesoscopic level organize their dynamics by alternating periods of absolute and
relative phase synchronization. Thus, neuron oscillating in the theta band can relative phase synchronization. Thus, neuron oscillating in the theta band can
be considered as metastable systems. The computation of the AS via the HT be considered as metastable systems. The computation of the AS via the HT
turns out to be a very useful approach. It allows us to compute the instantaneous turns out to be a very useful approach. It allows us to compute the instantaneous
phase and the instantaneous amplitude without assumptions about the linearity phase and the instantaneous amplitude without assumptions about the linearity
and stationary in the EEG records. The method is fairly robust since it is able to and stationary in the EEG records. The method is fairly robust since it is able to
capture the analytic frequencies in signals with well defined peak in the power capture the analytic frequencies in signals with well defined peak in the power
spectrum. The AS approach allows us to identify two regions in the up spectrum. The AS approach allows us to identify two regions in the up
unwrapped phase plot: Straight phases and scattered phases. Straight phases unwrapped phase plot: Straight phases and scattered phases. Straight phases
correspond to absolute phase synchronization while scattered phases correspond correspond to absolute phase synchronization while scattered phases correspond
to relative phase synchronization. to relative phase synchronization.

REFERENCES REFERENCES

Boashash, B. (1992). "Estimating and Interpreting the instantaneous frequency of a signal. I. Boashash, B. (1992). "Estimating and Interpreting the instantaneous frequency of a signal. I.
Fundamentals." Proceedings of the IEEE Vol 80(14): 520 - 538 Fundamentals." Proceedings of the IEEE Vol 80(14): 520 - 538
Crutchfield, J., D. Farmer, et al. (1980). "Power spectral analysis of dynamical system." Physics Crutchfield, J., D. Farmer, et al. (1980). "Power spectral analysis of dynamical system." Physics
Letters Vol 76 A (1). Letters Vol 76 A (1).
Freeman, W. (2004). "Origin, structure and role of the background EEG activity. Part 3. Neural Freeman, W. (2004). "Origin, structure and role of the background EEG activity. Part 3. Neural
frame classification." Clinical Neurophysilogy(116): 1118-1129. frame classification." Clinical Neurophysilogy(116): 1118-1129.
Freeman, W. (2006). Hilbert transform for brain waves. Berkeley University of California at Freeman, W. (2006). Hilbert transform for brain waves. Berkeley University of California at
Berkeley. Available on line at www.scholarpedia.org. Berkeley. Available on line at www.scholarpedia.org.
Kelso, S. (1995). Dynamic patterns: The self-organization of brain and behavior. Cambridge, MA, Kelso, S. (1995). Dynamic patterns: The self-organization of brain and behavior. Cambridge, MA,
MIT Press. MIT Press.
Losada, R. (2004). Practical FIR filter design in Matlab, The Math Works, Inc. Losada, R. (2004). Practical FIR filter design in Matlab, The Math Works, Inc.
Pikovsky, A., M. Rosenblum, et al. (1999). "Phase syncronization In regular and chaotic systems " Pikovsky, A., M. Rosenblum, et al. (1999). "Phase syncronization In regular and chaotic systems "
International Journal of Bifurcation and Chaos Vol 10(10): 2291-2305. International Journal of Bifurcation and Chaos Vol 10(10): 2291-2305.
Pikovsky, A., M. Rosenblum, et al. (2001). Synchronization. A universal concept in nonlinear Pikovsky, A., M. Rosenblum, et al. (2001). Synchronization. A universal concept in nonlinear
sciences, Cambridge University Press. sciences, Cambridge University Press.
Porat, B. (1997). A course in digital signal processing Jhon Wiley & Sons, Inc. Porat, B. (1997). A course in digital signal processing Jhon Wiley & Sons, Inc.
Rosenblum, M., A. Pikovsky, et al. (2001). Phase synchronization: From theory to data analysis. In Rosenblum, M., A. Pikovsky, et al. (2001). Phase synchronization: From theory to data analysis. In
F. Moss and S. Gielen, Editors, Handbook of biological physics. E. science. Vol 4: 297. F. Moss and S. Gielen, Editors, Handbook of biological physics. E. science. Vol 4: 297.
Strogatz, S. (1994). Nonlinear dynamics and chaos, Addison Wesley Reading. Strogatz, S. (1994). Nonlinear dynamics and chaos, Addison Wesley Reading.

6 6
14 14

REFERENCES REFERENCES

Downloaded From: http://ebooks.asmedigitalcollection.asme.org/ on 01/28/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

1 1
15 15

SIMULATION OF CELL COLLISION AND AGGREGATION USING A SIMULATION OF CELL COLLISION AND AGGREGATION USING A
THREE-DIMENSIONAL COMPUTATIONAL MODEL FOR THREE-DIMENSIONAL COMPUTATIONAL MODEL FOR
MULTICELLULAR TISSUE GROWTH MULTICELLULAR TISSUE GROWTH

BELGACEM BEN YOUSSEF LENNY TANG BELGACEM BEN YOUSSEF LENNY TANG
School of Interactive Arts & School of Interactive Arts & School of Interactive Arts & School of Interactive Arts &
Technology Technology Technology Technology
Simon Fraser University Simon Fraser University Simon Fraser University Simon Fraser University
Surrey, British Columbia, Canada Surrey, British Columbia, Canada Surrey, British Columbia, Canada Surrey, British Columbia, Canada

ABSTRACT ABSTRACT
We developed a computational model to simulate the growth of multicellular We developed a computational model to simulate the growth of multicellular
tissues using a discrete approach based on cellular automata. The model allows tissues using a discrete approach based on cellular automata. The model allows
us to study the tissue growth rates and population dynamics of different us to study the tissue growth rates and population dynamics of different
populations of migrating and proliferating mammalian cells. Cell migration is populations of migrating and proliferating mammalian cells. Cell migration is
modeled using a discrete-time Markov chain approach. We report simulation modeled using a discrete-time Markov chain approach. We report simulation
results describing cell collision and aggregation for two cell populations each results describing cell collision and aggregation for two cell populations each
having its own division and motion characteristics. Heterotypic and homotypic having its own division and motion characteristics. Heterotypic and homotypic
cell-cell interactions play important roles in cell and tissue functions. The cell-cell interactions play important roles in cell and tissue functions. The
temporal evolution of the frequency of cell collision and aggregation and their temporal evolution of the frequency of cell collision and aggregation and their
relations to other variables that quantify the dynamics of cell populations can relations to other variables that quantify the dynamics of cell populations can
be predicted by this model for different cell seeding topologies and cell be predicted by this model for different cell seeding topologies and cell
heterogeneity ratios. heterogeneity ratios.

INTRODUCTION INTRODUCTION
Each year, millions of surgical procedures are performed to relieve patients Each year, millions of surgical procedures are performed to relieve patients
who are affected by tissue loss, due to burns, injuries, or organ failure. who are affected by tissue loss, due to burns, injuries, or organ failure.
Operations treating patients using tissue reconstruction and organ transplantation Operations treating patients using tissue reconstruction and organ transplantation
have been highly successful. However, the number of patients treated by these have been highly successful. However, the number of patients treated by these
therapies is small due to the limited number of donors available. The primary therapies is small due to the limited number of donors available. The primary
focus of tissue engineering is the growth of three-dimensional tissues with focus of tissue engineering is the growth of three-dimensional tissues with
proper structure and function. Tissue engineers combine knowledge from the proper structure and function. Tissue engineers combine knowledge from the
areas of biochemistry, medical sciences, and engineering to develop bioartificial areas of biochemistry, medical sciences, and engineering to develop bioartificial
tissue substitutes or to induce tissue remodelling in order to repair, replace, or tissue substitutes or to induce tissue remodelling in order to repair, replace, or
enhance tissue functions (Langer and Vacanti, 1993; Mooney and Mikos, 1999). enhance tissue functions (Langer and Vacanti, 1993; Mooney and Mikos, 1999).

Tissue growth is a complex process affected by many contributing factors Tissue growth is a complex process affected by many contributing factors
such as the type of cells, initial seeding densities, spatial distribution of the seed such as the type of cells, initial seeding densities, spatial distribution of the seed
cells and the culture conditions (Ratcliffe and Niklason, 2002). Scaffold cells and the culture conditions (Ratcliffe and Niklason, 2002). Scaffold
properties, cell activities like adhesion or migration, and external stimuli that properties, cell activities like adhesion or migration, and external stimuli that
modulate cellular functions are among the many factors that affect the growth modulate cellular functions are among the many factors that affect the growth
rate of tissues. Hence, the development of bio-artificial tissue substitutes rate of tissues. Hence, the development of bio-artificial tissue substitutes
involves extensive and time-consuming experimentation. The availability of involves extensive and time-consuming experimentation. The availability of
computational models with predictive abilities will greatly speed up progress in computational models with predictive abilities will greatly speed up progress in
this area. this area.

1 1
15 15

Downloaded From: http://ebooks.asmedigitalcollection.asme.org/ on 01/28/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

2 2
16 16

This research focuses on the development of computational models to This research focuses on the development of computational models to
simulate the growth of three-dimensional tissues consisting of more than one simulate the growth of three-dimensional tissues consisting of more than one
cell type. Our specific objective here is to use such models to study the cell type. Our specific objective here is to use such models to study the
dynamics of cell-cell interactions. The success of models that describe the dynamics of cell-cell interactions. The success of models that describe the
dynamic behaviour of cell populations, however, will depend on our ability to dynamic behaviour of cell populations, however, will depend on our ability to
accurately describe the collision and aggregation of different cell types in accurately describe the collision and aggregation of different cell types in
various environments. various environments.

RELATED WORK RELATED WORK

Various modeling approaches have been used to simulate the population Various modeling approaches have been used to simulate the population
dynamics of proliferating cells. Early attempts to model cell population growth dynamics of proliferating cells. Early attempts to model cell population growth
were limited to nonmotile cells and the study of contact inhibition phenomena were limited to nonmotile cells and the study of contact inhibition phenomena
on the proliferation of anchorage-dependant endothelial cells. These early on the proliferation of anchorage-dependant endothelial cells. These early
models considered nonmotile cells proliferating in two dimensions (Zygourakis models considered nonmotile cells proliferating in two dimensions (Zygourakis
et al., 1991), or on microcarriers (Hawboldt et al., 1994). In disregarding cell et al., 1991), or on microcarriers (Hawboldt et al., 1994). In disregarding cell
locomotion, the cell growth rates in these early models are restricted by the locomotion, the cell growth rates in these early models are restricted by the
effects of contact inhibition. Lee et al. (1995) showed the importance of cell effects of contact inhibition. Lee et al. (1995) showed the importance of cell
motility and cell-cell interaction in describing the cell proliferation rates. Any motility and cell-cell interaction in describing the cell proliferation rates. Any
comprehensive model for tissue growth must consider these processes and comprehensive model for tissue growth must consider these processes and
account for the growth factors that regulate their rates. Later, Ben Youssef account for the growth factors that regulate their rates. Later, Ben Youssef
(1999) developed the first three-dimensional cellular automata model for tissue (1999) developed the first three-dimensional cellular automata model for tissue
growth. This work also utilizes a Markov chain approach to model the growth. This work also utilizes a Markov chain approach to model the
trajectories of migrating cells and is focused primarily on the study of a single trajectories of migrating cells and is focused primarily on the study of a single
population of proliferating and migrating cells. Cheng et al. (2006) enhanced population of proliferating and migrating cells. Cheng et al. (2006) enhanced
this model by accounting for the slowing of cells resulting from cell aggregation this model by accounting for the slowing of cells resulting from cell aggregation
while using one cell type. The latter two works form the basis for this model. while using one cell type. The latter two works form the basis for this model.

MODELING OF CELL COLLISION AND AGGREGATION MODELING OF CELL COLLISION AND AGGREGATION
To model our biological system, we focused our attention on four cellular To model our biological system, we focused our attention on four cellular
processes: cell division, cell migration, cell collision, and cell aggregation. The processes: cell division, cell migration, cell collision, and cell aggregation. The
first two processes are described elsewhere (Cheng et al., 2006). The latter two first two processes are described elsewhere (Cheng et al., 2006). The latter two
are described in the next two sections, respectively. These four processes are are described in the next two sections, respectively. These four processes are
illustrated on Figure 1. Mother cell 1, at the top left of the cellular space, illustrated on Figure 1. Mother cell 1, at the top left of the cellular space,
migrates downwards, and divides into two daughter cells. These two daughter migrates downwards, and divides into two daughter cells. These two daughter
cells, in this scenario, move around until they encounter each other, at which cells, in this scenario, move around until they encounter each other, at which
time aggregation occurs. If one of these daughter cells moves upwards and time aggregation occurs. If one of these daughter cells moves upwards and
encounters a cell of a different type (type 2), then collision occurs. encounters a cell of a different type (type 2), then collision occurs.

Cell Collision Cell Collision

Mammalian cells move in a certain direction for some period of time and, Mammalian cells move in a certain direction for some period of time and,
then, they turn and migrate in another direction. Cell collision occurs when a then, they turn and migrate in another direction. Cell collision occurs when a
cell moving in a certain direction encounters another cell of a different type in cell moving in a certain direction encounters another cell of a different type in
its path. When a cell of one type collides with a cell of a different type, the cell its path. When a cell of one type collides with a cell of a different type, the cell
slows down for a period of time and then changes its direction. The duration of slows down for a period of time and then changes its direction. The duration of
time a cell stays in the stationary state after a cell collision is based on its type time a cell stays in the stationary state after a cell collision is based on its type
(Lee et al., 1994; Kouvroukoglou et al., 1998). Once the cell is ready to resume (Lee et al., 1994; Kouvroukoglou et al., 1998). Once the cell is ready to resume

2 2
16 16

RELATED WORK RELATED WORK

Cell Collision Cell Collision

Downloaded From: http://ebooks.asmedigitalcollection.asme.org/ on 01/28/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

3 3
17 17

its motion, it migrates in a new direction. The modeling steps used in the cell its motion, it migrates in a new direction. The modeling steps used in the cell
collision process are described below: collision process are described below:
x The cell stops for a number of steps. x The cell stops for a number of steps.
x The cell changes its direction and resumes motion. x The cell changes its direction and resumes motion.

Figure 1 – An illustration showing the persistent random walk of two Figure 1 – An illustration showing the persistent random walk of two
mother cells of different types (1 and 2) during their division cycle. The two mother cells of different types (1 and 2) during their division cycle. The two
daughter cells move away from each other and may aggregate or collide daughter cells move away from each other and may aggregate or collide
with other cells. with other cells.

Cell Aggregation Cell Aggregation

Cell aggregation is a feature of tissue formation that allows the binding of Cell aggregation is a feature of tissue formation that allows the binding of
cells of the same type. It is this specific grouping of cells that enables the tissue cells of the same type. It is this specific grouping of cells that enables the tissue
to perform its intended purpose. Cell aggregation is the combination of two to perform its intended purpose. Cell aggregation is the combination of two
cellular functions: cell-to-cell recognition and cell adhesion. The self- cellular functions: cell-to-cell recognition and cell adhesion. The self-
recognition quality lets cells identify cells of the same type. When cells of the recognition quality lets cells identify cells of the same type. When cells of the
same type encounter each other, they adhere to one another and form a cellular same type encounter each other, they adhere to one another and form a cellular
aggregate (Wolf, 1983). As more cells of the same cell type encounter the aggregate (Wolf, 1983). As more cells of the same cell type encounter the
cellular aggregate, the cellular aggregate becomes larger forming a cluster of cellular aggregate, the cellular aggregate becomes larger forming a cluster of
cells. The adhesion can be strengthened by aggregation factors that are cells. The adhesion can be strengthened by aggregation factors that are
sometimes secreted by the cells (Dyson, 1978). As a result of aggregation, the sometimes secreted by the cells (Dyson, 1978). As a result of aggregation, the
cell slows down, “sticks” to another cell of the same type, and changes its cell slows down, “sticks” to another cell of the same type, and changes its
direction of motion. The basic steps used to model the process of cell direction of motion. The basic steps used to model the process of cell
aggregation are listed below: aggregation are listed below:
x The two cells stop for a number of steps, thereby entering an x The two cells stop for a number of steps, thereby entering an
aggregation state. aggregation state.
x The two cells change their direction and resume locomotion. x The two cells change their direction and resume locomotion.

Cell State Information Cell State Information

To characterize a heterotypic cell-cell interaction, our model uses a collision To characterize a heterotypic cell-cell interaction, our model uses a collision
stationary state (denoted by l = 0). If a cell c collides with another cell of a stationary state (denoted by l = 0). If a cell c collides with another cell of a

3 3
17 17

Cell Aggregation Cell Aggregation

Cell State Information Cell State Information

Downloaded From: http://ebooks.asmedigitalcollection.asme.org/ on 01/28/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

4 4
18 18

different type, cell c stops moving and enters what is known as the collision different type, cell c stops moving and enters what is known as the collision
stationary state. It remains in that state for a defined period equal to its expected stationary state. It remains in that state for a defined period equal to its expected
waiting time, E(T0). When E(T0) expires, cell c can move away in a randomly waiting time, E(T0). When E(T0) expires, cell c can move away in a randomly
assigned direction. assigned direction.

In the case of a homotypic cell-cell interaction, we added a second In the case of a homotypic cell-cell interaction, we added a second
stationary state (denoted by l = 7). Two colliding cells of the same type will stationary state (denoted by l = 7). Two colliding cells of the same type will
enter the aggregation state and stick together for a defined period equal to E(T7). enter the aggregation state and stick together for a defined period equal to E(T7).
The value of E(T7) indicates the likelihood for the cells to form multicellular The value of E(T7) indicates the likelihood for the cells to form multicellular
aggregates. When a cell collides with another cell of the same type, that is aggregates. When a cell collides with another cell of the same type, that is
already in the aggregation state, then the first cell will enter the aggregation state already in the aggregation state, then the first cell will enter the aggregation state
and both cells will reset their persistence counters to E(T7)/ǻt. Once the waiting and both cells will reset their persistence counters to E(T7)/ǻt. Once the waiting
time E(T7) has expired, the two cells can move away in randomly assigned time E(T7) has expired, the two cells can move away in randomly assigned
directions. directions.

The state of a cell takes values from the following set of eight digit integer The state of a cell takes values from the following set of eight digit integer
numbers: : {klmnpqrs | k , l , m, n, p, q, r , and s } , where k is the cell numbers: : {klmnpqrs | k , l , m, n, p, q, r , and s } , where k is the cell
type. The direction of motion is identified by the direction index l . When l is type. The direction of motion is identified by the direction index l . When l is
equal to 0, the cell is in the collision stationary state. When the value of l equal to 0, the cell is in the collision stationary state. When the value of l
ranges from 1 - 6, it represents one of six directions the cell is currently moving ranges from 1 - 6, it represents one of six directions the cell is currently moving
in (east, north, west, south, above, and below). When the value of l is 7, it in (east, north, west, south, above, and below). When the value of l is 7, it
enters an aggregation stationary state where it “sticks” to another cell of the enters an aggregation stationary state where it “sticks” to another cell of the
same type potentially forming cellular aggregates. The digits mn denote the same type potentially forming cellular aggregates. The digits mn denote the
persistence counter. It is the time remaining until the next change in the persistence counter. It is the time remaining until the next change in the
direction of cell movement. The cell phase counter is represented by the direction of cell movement. The cell phase counter is represented by the
digits pqrs . The cell phase counter is the time remaining before the cell digits pqrs . The cell phase counter is the time remaining before the cell
divides. divides.

Initial Conditions Initial Conditions

Two cell populations are used to study collision and aggregation. Each of Two cell populations are used to study collision and aggregation. Each of
these two cell populations, referred to as cell population 1 and cell population 2, these two cell populations, referred to as cell population 1 and cell population 2,
has its own division and migration chacteristics. The division time distributions has its own division and migration chacteristics. The division time distributions
for these two cell populations are given in Table 1. For instance, we observe for these two cell populations are given in Table 1. For instance, we observe
that for cell population 1, 64% of the living cells have division times between 12 that for cell population 1, 64% of the living cells have division times between 12
and 18 hours. Moreover, cells from population 1 are fast-moving cells that and 18 hours. Moreover, cells from population 1 are fast-moving cells that
migrate at 10 ȝm/hr while cells from population 2 are slow-moving cells migrate at 10 ȝm/hr while cells from population 2 are slow-moving cells
migrating at a speed of only 1 ȝm/hr. migrating at a speed of only 1 ȝm/hr.

Table 1 - Cell Division Time Distributions for the Two Cell Populations Table 1 - Cell Division Time Distributions for the Two Cell Populations
Cell Populations Cell Populations
Division Times (hrs) Cell Population 1 Cell Population 2 Division Times (hrs) Cell Population 1 Cell Population 2
[12,18) 64% 4% [12,18) 64% 4%
[18,24) 32% 32% [18,24) 32% 32%
[24,30) 4% 64% [24,30) 4% 64%

4 4
18 18

Initial Conditions Initial Conditions

Downloaded From: http://ebooks.asmedigitalcollection.asme.org/ on 01/28/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

5 5
19 19

Cell Seeding Distributions Cell Seeding Distributions

The uniform cell seeding topolgy is considered here. This topology was The uniform cell seeding topolgy is considered here. This topology was
chosen for its applicability to experimental research conducted in laboratories. chosen for its applicability to experimental research conducted in laboratories.
In this case, cells are randomly seeded in the cellular space. The uniform In this case, cells are randomly seeded in the cellular space. The uniform
topology simulates the migration and proliferation of cells in a sparsely and topology simulates the migration and proliferation of cells in a sparsely and
uniformly seeded environment with the objective of simulating tissue uniformly seeded environment with the objective of simulating tissue
regeneration. We have associated two types of cell distributions with this regeneration. We have associated two types of cell distributions with this
topology, a segmented distribution and a mixed one. topology, a segmented distribution and a mixed one.
In the segmented distribution, each cell type is seeded in a separate area of In the segmented distribution, each cell type is seeded in a separate area of
the cellular space. During the simulation, cells can migrate freely in the cellular the cellular space. During the simulation, cells can migrate freely in the cellular
space, and can enter areas that were originally designated for a particular type of space, and can enter areas that were originally designated for a particular type of
cell during the initial cell seeding. In the mixed distribution, the different cell cell during the initial cell seeding. In the mixed distribution, the different cell
types are seeded together in a random order according to the topology of cell types are seeded together in a random order according to the topology of cell
seeding. Figure 2 shows an illustration of these two cell distributions. In this seeding. Figure 2 shows an illustration of these two cell distributions. In this
topology, the seeding density indicates the percentage of sites occupied by cells topology, the seeding density indicates the percentage of sites occupied by cells
at the start of the simulation. at the start of the simulation.

(a) (b) (a) (b)

Figure 2 – A uniform seeding topology using two cell populations in (a) a Figure 2 – A uniform seeding topology using two cell populations in (a) a
segmented and (b) a mixed distribution. In each case, a total seeding segmented and (b) a mixed distribution. In each case, a total seeding
density of 0.5% is used. This example is based on a 20×20×20 cellular array. density of 0.5% is used. This example is based on a 20×20×20 cellular array.

Dynamics of Cell-Cell Interactions Dynamics of Cell-Cell Interactions

At the start of each simulation, the cellular space is seeded with a total At the start of each simulation, the cellular space is seeded with a total
number of seed cells equal to N0. These seed cells migrate and proliferate in the number of seed cells equal to N0. These seed cells migrate and proliferate in the
cellular array based on the cellular automata rules to simulate the dynamic cellular array based on the cellular automata rules to simulate the dynamic
process of tissue growth. After some time t, Nc(t) sites of the cellular automaton process of tissue growth. After some time t, Nc(t) sites of the cellular automaton
are occupied by cells. In addition, the frequency of cell-cell interactions, Fc (t ) , are occupied by cells. In addition, the frequency of cell-cell interactions, Fc (t ) ,
which reflects the total number of interactions for a given cell per simulation which reflects the total number of interactions for a given cell per simulation
step 't , is provided by Equation (4). Thus, Fc (t ) yields the sum of the step 't , is provided by Equation (4). Thus, Fc (t ) yields the sum of the
number of cell collisions, or heterotypic cell interactions, and the number of cell number of cell collisions, or heterotypic cell interactions, and the number of cell
aggregations, also known as homotypic cell interactions. aggregations, also known as homotypic cell interactions.
Total number of cell-cell interactions in interval [t, t+ ǻt] Total number of cell-cell interactions in interval [t, t+ ǻt]
Fc (t ) (4) Fc (t ) (4)
Nc(t) × ǻt Nc(t) × ǻt

5 5
19 19

Cell Seeding Distributions Cell Seeding Distributions

(a) (b) (a) (b)

Dynamics of Cell-Cell Interactions Dynamics of Cell-Cell Interactions

Downloaded From: http://ebooks.asmedigitalcollection.asme.org/ on 01/28/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

6 6
20 20

We also define the heterogeneity measure, H, as the ratio of the initially We also define the heterogeneity measure, H, as the ratio of the initially
seeded number of cells from population 1 to that from population 2. The ratio H seeded number of cells from population 1 to that from population 2. The ratio H
is given by: is given by:
initial number of cells from population 1 initial number of cells from population 1
H . (5) H . (5)
initial number of cells from population 2 initial number of cells from population 2
That is, when H = 9, there are 9 cells from population 1 for every cell from That is, when H = 9, there are 9 cells from population 1 for every cell from
population 2 in the cell seeding density. population 2 in the cell seeding density.

SIMULATION RESULTS AND DISCUSSION SIMULATION RESULTS AND DISCUSSION

The presented simulation results were obtained using a 200200200 The presented simulation results were obtained using a 200200200
cellular array, a 99.99% confluence parameter, and average waiting times of 2 cellular array, a 99.99% confluence parameter, and average waiting times of 2
hoursfor the six directional states and 1 hour each for the aggregation state and hoursfor the six directional states and 1 hour each for the aggregation state and
the collision state, respectively. the collision state, respectively.

Segmented Uniform Distribution Segmented Uniform Distribution

The frequency of cell-cell interactions versus time for both cell population 1 and The frequency of cell-cell interactions versus time for both cell population 1 and
cell population 2 for various values of H is shown in Figure 3. For cell cell population 2 for various values of H is shown in Figure 3. For cell
population 1, as H is increased the frequency of cell-cell interactions increases population 1, as H is increased the frequency of cell-cell interactions increases
and reaches a higher peak value due to an increase in the number of fast cells in and reaches a higher peak value due to an increase in the number of fast cells in
its segment of the cellular space, which leads to more cell-to-cell interactions. its segment of the cellular space, which leads to more cell-to-cell interactions.
Because cells in population 2 have a lower speed of 1 ȝm/hr, they tend to form Because cells in population 2 have a lower speed of 1 ȝm/hr, they tend to form
more localized clusters. When combined with the fact that an increase in the more localized clusters. When combined with the fact that an increase in the
ratio H also reduces the size of the segment for this population (and the number ratio H also reduces the size of the segment for this population (and the number
of slow cells), this results in fewer and fewer cell-cell interactions as H is of slow cells), this results in fewer and fewer cell-cell interactions as H is
increased. increased.

Figure 3 - The effect of varying the ratio H on the frequency of cell-cell Figure 3 - The effect of varying the ratio H on the frequency of cell-cell
interactions for cell population 1 (left) and cell population 2 (right). interactions for cell population 1 (left) and cell population 2 (right).

Figures 4 displays the temporal evolution of the frequency of cell-cell Figures 4 displays the temporal evolution of the frequency of cell-cell
interactions for H 1 and H 9 . The homotypic and hetrotypic cell interactions for H 1 and H 9 . The homotypic and hetrotypic cell
interactions are also shown in both cases. We observe that cell collision interactions are also shown in both cases. We observe that cell collision
dominates early in the proliferation process. Cell aggregation becomes dominates early in the proliferation process. Cell aggregation becomes

6 6
20 20

SIMULATION RESULTS AND DISCUSSION SIMULATION RESULTS AND DISCUSSION

Segmented Uniform Distribution Segmented Uniform Distribution

Downloaded From: http://ebooks.asmedigitalcollection.asme.org/ on 01/28/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

7 7
21 21

important only later in the process, even as H is increased. These results are important only later in the process, even as H is increased. These results are
also confirmed for the mixed uniform distribution and are not be shown here. also confirmed for the mixed uniform distribution and are not be shown here.

Figure 4 - The temporal evolution of the frequency of cell-cell interactions Figure 4 - The temporal evolution of the frequency of cell-cell interactions
and its two components for cell population 1 when H=1 (left) and H=9 and its two components for cell population 1 when H=1 (left) and H=9
(right). (right).

Mixed Uniform Distribution Mixed Uniform Distribution

Figure 5 shows the frequency of cell-cell interactions for populations 1 and Figure 5 shows the frequency of cell-cell interactions for populations 1 and
2 for different values of H given a fixed cell seeding density of 0.5%. For cells 2 for different values of H given a fixed cell seeding density of 0.5%. For cells
in population 1, during the first two days and as H is increased, the frequency of in population 1, during the first two days and as H is increased, the frequency of
cell-cell interactions increases due to an increase in interactions between fast cell-cell interactions increases due to an increase in interactions between fast
and slow moving cells early in the proliferation process. This is facilitated by and slow moving cells early in the proliferation process. This is facilitated by
the nature of the mixed seeding distribution. the nature of the mixed seeding distribution.

Figure 5 - The effect of varying the ratio H on the frequency of cell-cell Figure 5 - The effect of varying the ratio H on the frequency of cell-cell
interactions for cell population 1 (left) and cell population 2 (right). interactions for cell population 1 (left) and cell population 2 (right).

This phenomenon is clearly illustrated in Figure 6. The figure depicts both This phenomenon is clearly illustrated in Figure 6. The figure depicts both
the frequency of cell aggregations as well as the frequency of cell collisions for the frequency of cell aggregations as well as the frequency of cell collisions for

7 7
21 21

Mixed Uniform Distribution Mixed Uniform Distribution

Downloaded From: http://ebooks.asmedigitalcollection.asme.org/ on 01/28/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

8 8
22 22

H =1 and H =9. After the first two days, the frequency of cell collisions H =1 and H =9. After the first two days, the frequency of cell collisions
decreases rapidly as H is increased. Given that the frequency of homotypic cell decreases rapidly as H is increased. Given that the frequency of homotypic cell
interactions between fast cells of the same type increases with time until it interactions between fast cells of the same type increases with time until it
reaches a maximum for all values of H, the combined effect of these two reaches a maximum for all values of H, the combined effect of these two
observations leads to a reversal of behaviour. That is, the frequency of cell-cell observations leads to a reversal of behaviour. That is, the frequency of cell-cell
interactions decreases as H is increased after the first two days for cells in interactions decreases as H is increased after the first two days for cells in
population 1. For cells in population 2, the frequency of cell-cell interactions population 1. For cells in population 2, the frequency of cell-cell interactions
decreases as H is increased due to the decrease in the seeded number of slow decreases as H is increased due to the decrease in the seeded number of slow
cells. This results in a smaller number of cell clusters, which in turn impacts cells. This results in a smaller number of cell clusters, which in turn impacts
negatively the number of homotypic cell interactions between slow cells. negatively the number of homotypic cell interactions between slow cells.

Figure 6 - The temporal evolution of the frequency of cell-cell interactions Figure 6 - The temporal evolution of the frequency of cell-cell interactions
and its two components for cell population 1 when H=1 (left) and H=9 and its two components for cell population 1 when H=1 (left) and H=9
(right). (right).

CONCLUSIONS CONCLUSIONS
In this paper, we presented simulation results for cell collision and cell In this paper, we presented simulation results for cell collision and cell
aggregation obtained using a three-dimensional computational model for aggregation obtained using a three-dimensional computational model for
multicellular tissue growth. When employing two types of cell populations with multicellular tissue growth. When employing two types of cell populations with
different migration speeds and division time distributions, our simulations different migration speeds and division time distributions, our simulations
showed that increasing the cell heterogeneity ratio H of fast to slow cells showed that increasing the cell heterogeneity ratio H of fast to slow cells
enhanced the frequency of cell-cell interactions in a uniform seeding topology. enhanced the frequency of cell-cell interactions in a uniform seeding topology.
It appears that the frequency of cell-cell interactions may depend at least It appears that the frequency of cell-cell interactions may depend at least
primarily on the dynamics of the cell populations present in the cellular space. primarily on the dynamics of the cell populations present in the cellular space.
Throughout these results, we noted that a rapid increase in the fraction of Throughout these results, we noted that a rapid increase in the fraction of
surrounded cells is accompanied by a fast decrease in the frequency of cell-cell surrounded cells is accompanied by a fast decrease in the frequency of cell-cell
interactions. Moreover, the segmented uniform distribution yielded a smaller interactions. Moreover, the segmented uniform distribution yielded a smaller
overall frequency of cell-cell interactions for the population of fast cells. In the overall frequency of cell-cell interactions for the population of fast cells. In the
case of slow-moving cells, however, nearly similar simulation results were case of slow-moving cells, however, nearly similar simulation results were
obtained by both cell seeding distributions. These conclusions may have obtained by both cell seeding distributions. These conclusions may have
important implications for the design of experiments that can test the efficacy of important implications for the design of experiments that can test the efficacy of
cell heterogeneity and cell seeding distributions designed to enhance or cell heterogeneity and cell seeding distributions designed to enhance or
minimize the frequency of cell-cell interactions in a tissue growth environment. minimize the frequency of cell-cell interactions in a tissue growth environment.

8 8
22 22

Downloaded From: http://ebooks.asmedigitalcollection.asme.org/ on 01/28/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

9 9
23 23

To achieve the desired goals, it is recommended that assays based on their To achieve the desired goals, it is recommended that assays based on their
corresponding design factors be adopted. corresponding design factors be adopted.

REFERENCES REFERENCES

Ben Youssef, B., 1999, Cell proliferation and migration: 3-D modelling using cellular automata, Ben Youssef, B., 1999, Cell proliferation and migration: 3-D modelling using cellular automata,
development of a parallel algorithm and its parallel implementation on an IBM SP2, Ph.D. development of a parallel algorithm and its parallel implementation on an IBM SP2, Ph.D.
Dissertation, University of Houston. Dissertation, University of Houston.

Ben Youssef, B., Cheng, G., Zygourakis, K. and Markenscoff, P., 2007, Parallel implementation of a Ben Youssef, B., Cheng, G., Zygourakis, K. and Markenscoff, P., 2007, Parallel implementation of a
cellular automaton modeling the growth of three-dimensional tissues. International Journal of cellular automaton modeling the growth of three-dimensional tissues. International Journal of
High Performance Computing Applications, Vol. 21, No. 2, pp. 196-209. High Performance Computing Applications, Vol. 21, No. 2, pp. 196-209.

Cheng, G., Ben Youssef, B., Markenscoff, P. and Zygourakis, K., 2006, Cell population dynamics Cheng, G., Ben Youssef, B., Markenscoff, P. and Zygourakis, K., 2006, Cell population dynamics
modulate the rates of tissue growth processes. Biophysical journal, Vol. 90, No. 3, pp. 713- modulate the rates of tissue growth processes. Biophysical journal, Vol. 90, No. 3, pp. 713-
724. 724.

Dyson, R., 1978, Essentials of cell biology, Allyn and Bacon, Inc., Boston, MA. Dyson, R., 1978, Essentials of cell biology, Allyn and Bacon, Inc., Boston, MA.

Hawboldt, K.A., Kalogerakis, N. and Behie, L.A., 1994, A cellular automaton model for Hawboldt, K.A., Kalogerakis, N. and Behie, L.A., 1994, A cellular automaton model for
microcarrier cultures. Biotechnology and bioengineering, Vol. 43, No. 1, pp. 90-100. microcarrier cultures. Biotechnology and bioengineering, Vol. 43, No. 1, pp. 90-100.

Kouvroukoglou, S., Lakkis, C., Wallace, D., Zygourakis, K. and Epner, D., 1998, Bioenergetics of Kouvroukoglou, S., Lakkis, C., Wallace, D., Zygourakis, K. and Epner, D., 1998, Bioenergetics of
rat prostate cancer cell migration. The Prostate, Vol. 34, No. 2, pp. 137-144. rat prostate cancer cell migration. The Prostate, Vol. 34, No. 2, pp. 137-144.

Langer, R. and Vacanti, J.P., 1993, Tissue engineering. Science, Vol. 260, pp. 920-926. Langer, R. and Vacanti, J.P., 1993, Tissue engineering. Science, Vol. 260, pp. 920-926.

Lee, Y., Kouvroukoglou, S., McIntire, L.V. and Zygourakis, K., 1995, A cellular automaton model Lee, Y., Kouvroukoglou, S., McIntire, L.V. and Zygourakis, K., 1995, A cellular automaton model
for the proliferation of migrating contact-inhibited cells. Biophysical journal, Vol. 69, No. 10, for the proliferation of migrating contact-inhibited cells. Biophysical journal, Vol. 69, No. 10,
pp. 1284-1298. pp. 1284-1298.

Lee, Y., McIntire, L.V. and Zygourakis, K., 1994, Analysis of endothelial cell locomotion: Lee, Y., McIntire, L.V. and Zygourakis, K., 1994, Analysis of endothelial cell locomotion:
Differential effects of motility and contact inhibition. Biotechnology and bioengineering, Vol. Differential effects of motility and contact inhibition. Biotechnology and bioengineering, Vol.
43, No. 7, pp. 622-634. 43, No. 7, pp. 622-634.

Marée, A.F. and Hogeweg, P., 2001, How amoeboids self-organize into a fruiting body: Marée, A.F. and Hogeweg, P., 2001, How amoeboids self-organize into a fruiting body:
Multicellular coordination in dictyostelium discoideum. Proceedings of the National Academy Multicellular coordination in dictyostelium discoideum. Proceedings of the National Academy
of Sciences, Vol. 98, No. 7, pp. 3879-3883. of Sciences, Vol. 98, No. 7, pp. 3879-3883.

Mooney, D.J. and Mikos, A.G., 1999, Growing new organs. Scientific American, Vol. 280, pp. 60- Mooney, D.J. and Mikos, A.G., 1999, Growing new organs. Scientific American, Vol. 280, pp. 60-
65. 65.

Ratcliffe, A. and Niklason, L.E., 2002, Bioreactors and bioprocessing for tissue engineering. Annals Ratcliffe, A. and Niklason, L.E., 2002, Bioreactors and bioprocessing for tissue engineering. Annals
of the New York Academy of Sciences, Vol. 961, No. 1, pp. 210-215. of the New York Academy of Sciences, Vol. 961, No. 1, pp. 210-215.

Wolfe, S., 1983, Introduction to cell biology, Wadsworth Publishing Co., Belmont, CA. Wolfe, S., 1983, Introduction to cell biology, Wadsworth Publishing Co., Belmont, CA.

Zygourakis, K., Bizios, R. and Markenscoff, P., 1991, Proliferation of anchorage-dependent contact- Zygourakis, K., Bizios, R. and Markenscoff, P., 1991, Proliferation of anchorage-dependent contact-
inhibited cells: I. development of theoretical models based on cellular automata. Biotechnology inhibited cells: I. development of theoretical models based on cellular automata. Biotechnology
and bioengineering, Vol. 38, No. 5, pp. 459-470. and bioengineering, Vol. 38, No. 5, pp. 459-470.

9 9
23 23

REFERENCES REFERENCES

Dyson, R., 1978, Essentials of cell biology, Allyn and Bacon, Inc., Boston, MA. Dyson, R., 1978, Essentials of cell biology, Allyn and Bacon, Inc., Boston, MA.

Langer, R. and Vacanti, J.P., 1993, Tissue engineering. Science, Vol. 260, pp. 920-926. Langer, R. and Vacanti, J.P., 1993, Tissue engineering. Science, Vol. 260, pp. 920-926.

Mooney, D.J. and Mikos, A.G., 1999, Growing new organs. Scientific American, Vol. 280, pp. 60- Mooney, D.J. and Mikos, A.G., 1999, Growing new organs. Scientific American, Vol. 280, pp. 60-
65. 65.

Wolfe, S., 1983, Introduction to cell biology, Wadsworth Publishing Co., Belmont, CA. Wolfe, S., 1983, Introduction to cell biology, Wadsworth Publishing Co., Belmont, CA.

Downloaded From: http://ebooks.asmedigitalcollection.asme.org/ on 01/28/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

1 1
25 25

CYSTEINES ON AMINO TERMINUS INFER OXIDATION STATES OF CYSTEINES ON AMINO TERMINUS INFER OXIDATION STATES OF
OTHER CYSTEINES ON PROTEIN CHAINS OTHER CYSTEINES ON PROTEIN CHAINS

AIGUO DU YI PAN AIGUO DU YI PAN

Georgia State University Georgia State University Georgia State University Georgia State University
34 Peachtree Street, Suite 1450 34 Peachtree Street, Suite 1450 34 Peachtree Street, Suite 1450 34 Peachtree Street, Suite 1450
Atlanta, Georgia Atlanta, Georgia Atlanta, Georgia Atlanta, Georgia

ABSTRACT ABSTRACT
The binding states of cysteine residues on protein carry important information The binding states of cysteine residues on protein carry important information
about the function and structure of proteins. In this research, support vector about the function and structure of proteins. In this research, support vector
machine (SVM) was used for the prediction of bonding states of cysteines. We machine (SVM) was used for the prediction of bonding states of cysteines. We
found that the oxidation states of amino terminus cysteines infer the oxidation found that the oxidation states of amino terminus cysteines infer the oxidation
states of other cysteines in the same protein chain. By including the oxidations states of other cysteines in the same protein chain. By including the oxidations
states of up to two cysteines on the amino terminus and the flanking sequences states of up to two cysteines on the amino terminus and the flanking sequences
around the cysteine as well as the global protein information, accuracy of 95% around the cysteine as well as the global protein information, accuracy of 95%
is achieved for the data set consisting of 4136 cysteine-containing segments is achieved for the data set consisting of 4136 cysteine-containing segments
extracted from 969 nonhomologous proteins. extracted from 969 nonhomologous proteins.

INTRODUCTION INTRODUCTION
Cysteine is one of the few amino acids that contain sulfur. This allows cysteine to bond in Cysteine is one of the few amino acids that contain sulfur. This allows cysteine to bond in
a special way though disulfide bond and help to maintain/stabilize the 3-D structure of a special way though disulfide bond and help to maintain/stabilize the 3-D structure of
proteins. In addition, cysteines play very important roles in the function, properties as proteins. In addition, cysteines play very important roles in the function, properties as
well as aging process of proteins. As is illustrated in Figure 1, cysteine residues have two well as aging process of proteins. As is illustrated in Figure 1, cysteine residues have two
possible chemical states in proteins: oxidized state and reduced state. These two states are possible chemical states in proteins: oxidized state and reduced state. These two states are
interchangeable when proper condition is met. In its reduced form, cysteine can undergo interchangeable when proper condition is met. In its reduced form, cysteine can undergo
chemical reactions such as alkylation, oxidation or forming complex compounds with chemical reactions such as alkylation, oxidation or forming complex compounds with
metal ions. These chemical reactions carry critical biological meanings such as metal ions. These chemical reactions carry critical biological meanings such as
activation, deactivation of the active sites of enzymes and changes in the local activation, deactivation of the active sites of enzymes and changes in the local
environment of the protein. In their oxidized form, two cysteine residues from the same environment of the protein. In their oxidized form, two cysteine residues from the same
protein chain form intra-chain disulfide bond, which links distant portion of a protein protein chain form intra-chain disulfide bond, which links distant portion of a protein
chain and provides strong structural constraints in the form of long-range interactions. chain and provides strong structural constraints in the form of long-range interactions.
Two cysteines from different proteins form inter-chain disulfide bond and enable more Two cysteines from different proteins form inter-chain disulfide bond and enable more
complicated protein structures and functions. Interchange between its reduced form and complicated protein structures and functions. Interchange between its reduced form and
oxidized form also serve as regulation switches for enzymes and other proteins [1,2]. oxidized form also serve as regulation switches for enzymes and other proteins [1,2].
Therefore, accurate prediction of the oxidation states of cysteine is essential to the studies Therefore, accurate prediction of the oxidation states of cysteine is essential to the studies
of protein stability, protein function and three-dimensional structure[3~6] of proteins. of protein stability, protein function and three-dimensional structure[3~6] of proteins.

1 1
25 25

CYSTEINES ON AMINO TERMINUS INFER OXIDATION STATES OF CYSTEINES ON AMINO TERMINUS INFER OXIDATION STATES OF
OTHER CYSTEINES ON PROTEIN CHAINS OTHER CYSTEINES ON PROTEIN CHAINS

AIGUO DU YI PAN AIGUO DU YI PAN

Downloaded From: http://ebooks.asmedigitalcollection.asme.org/ on 01/28/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

2 2
26 26

Fig 1: oxidation states of cysteine Fig 1: oxidation states of cysteine

There have been a number of studies in the prediction of the bonding states of cysteines. There have been a number of studies in the prediction of the bonding states of cysteines.
Muskal et al [7] obtained 81% accuracy by using neural network and the sequence Muskal et al [7] obtained 81% accuracy by using neural network and the sequence
information around the cysteines. Fiser et al.[8] performed statistical analysis on the information around the cysteines. Fiser et al.[8] performed statistical analysis on the
amino acid frequencies in the sequence environment of cysteines and achieved 71% amino acid frequencies in the sequence environment of cysteines and achieved 71%
accuracy. Mucchielli-Giorgi et al.[9] assessed the relative efficiency of different accuracy. Mucchielli-Giorgi et al.[9] assessed the relative efficiency of different
descriptors in predicting the cysteine disulfide bonding states and concluded that the descriptors in predicting the cysteine disulfide bonding states and concluded that the
amino acid content of the whole protein is more informative than the flanking sequences. amino acid content of the whole protein is more informative than the flanking sequences.
This approach has an accuracy of 84%. Martelli et al.[10,11] implemented a new hybrid This approach has an accuracy of 84%. Martelli et al.[10,11] implemented a new hybrid
system that combines a neural network and a hidden Markov model (hidden neural system that combines a neural network and a hidden Markov model (hidden neural
network) and obtained an overall accuracy of 88%. Frasconi et al.[12] introduced a network) and obtained an overall accuracy of 88%. Frasconi et al.[12] introduced a
support vector machines (SVM) based predictor which operates in two stages and support vector machines (SVM) based predictor which operates in two stages and
incorporated both information at the protein level and local sequences. The prediction incorporated both information at the protein level and local sequences. The prediction
accuracy of this approach is 83.6%. By using the SVM based on multiple feature vectors accuracy of this approach is 83.6%. By using the SVM based on multiple feature vectors
and assuming that the transition probability of cysteines’ oxidation states in the testing and assuming that the transition probability of cysteines’ oxidation states in the testing
dataset was known beforehand, Chen et al.[13] achieved the record high accuracy of dataset was known beforehand, Chen et al.[13] achieved the record high accuracy of
90%. They pointed out that cysteine state sequence in a protein chain has significant 90%. They pointed out that cysteine state sequence in a protein chain has significant
influence on the prediction accuracy. Most recently, Ceroni et al. [14] employed SVM influence on the prediction accuracy. Most recently, Ceroni et al. [14] employed SVM
binary classifier to predict the bonding states of each cysteine, followed by a refinement binary classifier to predict the bonding states of each cysteine, followed by a refinement
step that take the overall bonding state assignment of the entire chain into consideration step that take the overall bonding state assignment of the entire chain into consideration
and reported an accuracy of 88%. and reported an accuracy of 88%.
In this research, we found that the oxidation states of amino terminal cysteines infer the In this research, we found that the oxidation states of amino terminal cysteines infer the
oxidation states of other cysteines in the same protein chain. The encodings includes the oxidation states of other cysteines in the same protein chain. The encodings includes the
ordinal number (the order of the cysteine in the whole protein chain in comparison to ordinal number (the order of the cysteine in the whole protein chain in comparison to
other cysteines) of the cysteine in consideration, the oxidation states pattern of up to two other cysteines) of the cysteine in consideration, the oxidation states pattern of up to two
amino terminal cysteines on the protein as well as the flanking sequence information and amino terminal cysteines on the protein as well as the flanking sequence information and
global information of protein chain. It was found that SVM based on this set of global information of protein chain. It was found that SVM based on this set of
encodings gave improved overall accuracy (94.8%) for the data set consisting of 4136 encodings gave improved overall accuracy (94.8%) for the data set consisting of 4136
cysteine-containing segments extracted from 969 nonhomologous proteins. This is 4.8% cysteine-containing segments extracted from 969 nonhomologous proteins. This is 4.8%
higher than the current record of 90% by Chen et al.[13]. The influence of the local higher than the current record of 90% by Chen et al.[13]. The influence of the local
flanking sequence of cysteines, the global information such as the number of cysteines flanking sequence of cysteines, the global information such as the number of cysteines
present and amino acids composition of the whole protein chain were also studied. present and amino acids composition of the whole protein chain were also studied.

METHOD METHOD
In this work, SVM were used to predict the potential bonding state of cysteines. Support In this work, SVM were used to predict the potential bonding state of cysteines. Support
Vector Machine (SVM) is a learning system that uses a hypothesis space of linear Vector Machine (SVM) is a learning system that uses a hypothesis space of linear
functions in a high dimensional feature space, trained with a learning algorithm from functions in a high dimensional feature space, trained with a learning algorithm from
optimization theory [15]. This learning scheme was designed by Vapnik and his co- optimization theory [15]. This learning scheme was designed by Vapnik and his co-
workers [16] and proved to be very powerful machine learning technique since its workers [16] and proved to be very powerful machine learning technique since its
introduction. Compared with other machine learning techniques, SVM has a number of introduction. Compared with other machine learning techniques, SVM has a number of
superior properties, such as effective avoidance of over-fitting, the ability to handle large superior properties, such as effective avoidance of over-fitting, the ability to handle large

2 2
26 26

Fig 1: oxidation states of cysteine Fig 1: oxidation states of cysteine

Downloaded From: http://ebooks.asmedigitalcollection.asme.org/ on 01/28/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

3 3
27 27

feature spaces and information condensing of the given data set etc. It has been feature spaces and information condensing of the given data set etc. It has been
successfully applied to a wide range of pattern recognition problems, including isolated successfully applied to a wide range of pattern recognition problems, including isolated
handwritten digit recognition, object recognition, speaker identification, and text handwritten digit recognition, object recognition, speaker identification, and text
categorization, etc. categorization, etc.
Given data x1, . . . , x1, the label for these data was set as +1 if xi is in class 1 (in Given data x1, . . . , x1, the label for these data was set as +1 if xi is in class 1 (in
our case, +1 if the cysteine residue is intra-chain bounded with another cysteine) and as - our case, +1 if the cysteine residue is intra-chain bounded with another cysteine) and as -
1 if xi belongs to class 2 (in our case, -1 if the cysteine residue is non-bounded or is inter- 1 if xi belongs to class 2 (in our case, -1 if the cysteine residue is non-bounded or is inter-
chain bounded with other cysteines). Then with these training data, SVM solves an chain bounded with other cysteines). Then with these training data, SVM solves an
optimization problem for binary classification: optimization problem for binary classification:

1 T l
(1) 1 T l
(1)
min Z Z C ¦[ min Z Z C ¦[
Z , b ,[ 2 i 1
i Z , b ,[ 2 i 1
i

and y (Z M ( x ) b) t 1 [ , [
T
t0 (2) and y (Z M ( x ) b) t 1 [ , [
T
t0 (2)
i i i i i i i i

Where xi is mapped to a higher dimensional space by the functionM; [i is the Where xi is mapped to a higher dimensional space by the functionM; [i is the
allowable training error; C is the cost of error. The software SVMlight [17] was used in our allowable training error; C is the cost of error. The software SVMlight [17] was used in our
experiments. experiments.
We first train an SVM to recognize the bounded and unbounded cysteines. Based We first train an SVM to recognize the bounded and unbounded cysteines. Based
on the training dataset, SVMlight returns a value regarding the bonding state of the on the training dataset, SVMlight returns a value regarding the bonding state of the
cysteines. The decision was made based on the value returned by SVM. The cysteine is cysteines. The decision was made based on the value returned by SVM. The cysteine is
considered intra-chain bounded if the returned value was larger than 0. Otherwise, the considered intra-chain bounded if the returned value was larger than 0. Otherwise, the
cysteine was treated as non-bounded or inter-chain bounded. cysteine was treated as non-bounded or inter-chain bounded.
Multiple feature vectors composed of information on flanking sequences, global Multiple feature vectors composed of information on flanking sequences, global
information of the protein chain, oxidation states pattern of cysteines and ordinal number information of the protein chain, oxidation states pattern of cysteines and ordinal number
of the cysteine being considered were used in this study. We followed the previous of the cysteine being considered were used in this study. We followed the previous
literatures [7~14] in that only those cysteines that form intra-chain disulfide bonds were literatures [7~14] in that only those cysteines that form intra-chain disulfide bonds were
treated as in oxidized states and all free cysteines and those that form inter-chain disulfide treated as in oxidized states and all free cysteines and those that form inter-chain disulfide
bonds are treated as in reduced form. The global information vectors consist of two set of bonds are treated as in reduced form. The global information vectors consist of two set of
vectors. One corresponds to the total number of cysteines and the total number of amino vectors. One corresponds to the total number of cysteines and the total number of amino
acids on the chain. The other corresponds to the amino acids composition (calculated by acids on the chain. The other corresponds to the amino acids composition (calculated by
number of particular amino acid/total number of amino acid). The ordinal number vector number of particular amino acid/total number of amino acid). The ordinal number vector
denotes where the cysteine is located relative to the other cysteines. The flanking denotes where the cysteine is located relative to the other cysteines. The flanking
sequence vectors are obtained from the adjacent amino acids of the interested cysteines. sequence vectors are obtained from the adjacent amino acids of the interested cysteines.
Representations developed by Meiler et al. [18] were used in this study, which accounted Representations developed by Meiler et al. [18] were used in this study, which accounted
for graph shape index, polarizability, volume, hydrophobicity and isoelectric point of for graph shape index, polarizability, volume, hydrophobicity and isoelectric point of
amino acids. Finally the vectors for oxidation state pattern of up to two amino terminus amino acids. Finally the vectors for oxidation state pattern of up to two amino terminus
cysteines were used. When it is the first or second cysteine’s oxidations state is to be cysteines were used. When it is the first or second cysteine’s oxidations state is to be
predicted, then the oxidation state of that cysteine is not encoded. predicted, then the oxidation state of that cysteine is not encoded.
The dataset in Martelli et al. [9] was used for testing purposes in this study. The The dataset in Martelli et al. [9] was used for testing purposes in this study. The
dataset is comprised of 4136 cysteine-containing segments. There are 1446 segments in dataset is comprised of 4136 cysteine-containing segments. There are 1446 segments in
the disulfide bonded states and 2690 segments in the non-bonded states. The segments the disulfide bonded states and 2690 segments in the non-bonded states. The segments
were extracted from 969 non homologous proteins from Protein Data Bank. The were extracted from 969 non homologous proteins from Protein Data Bank. The
sequence identity of the set is less than 25%. sequence identity of the set is less than 25%.
The prediction accuracy was calculated by following the standard conventions[19]: The prediction accuracy was calculated by following the standard conventions[19]:
Accuracy for prediction is calculated by: Accuracy for prediction is calculated by:
Q2 = Nc/N0 (3) Q2 = Nc/N0 (3)
where Nc is the total number of correctly predicted cysteines and N0 is the total number where Nc is the total number of correctly predicted cysteines and N0 is the total number
of cysteines. of cysteines.

3 3
27 27

1 T l
(1) 1 T l
(1)
C ¦[ C ¦[
2Z Z 2Z Z
min min
Z , b ,[ i Z , b ,[ i
i 1 i 1

and y (Z M ( x ) b) t 1 [ , [
T
t0 (2) and y (Z M ( x ) b) t 1 [ , [
T
t0 (2)
i i i i i i i i

Downloaded From: http://ebooks.asmedigitalcollection.asme.org/ on 01/28/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

4 4
28 28

Specificity is calculated by: Specificity is calculated by:

TN x TN x
Spec = (4) Spec = (4)
TN x FPx TN x FPx

And sensitivity is calculated by: And sensitivity is calculated by:

TPx TPx
Sens. = (5) Sens. = (5)
TPx FN x TPx FN x
where x denotes the bonded cysteines or non-bonded cysteines, FPx is the number of false where x denotes the bonded cysteines or non-bonded cysteines, FPx is the number of false
negatives in the prediction and TNx is the number of true negative predictions for negatives in the prediction and TNx is the number of true negative predictions for
bonding state x, FNx is the number of false negatives for bonding state x. and TPx is the bonding state x, FNx is the number of false negatives for bonding state x. and TPx is the
number of true negative predictions for bonding state x number of true negative predictions for bonding state x

The Matthews correlation coefficient (MCC) [20] is calculated as: The Matthews correlation coefficient (MCC) [20] is calculated as:

MCC= TPx TN x FPx FN x (6) MCC= TPx TN x FPx FN x (6)

(TPx FN x )(TPx FPx )(TN x FPx )(TN x FN x ) (TPx FN x )(TPx FPx )(TN x FPx )(TN x FN x )

The value of MCC is an indication of how good is the prediction. The closer the MCC is The value of MCC is an indication of how good is the prediction. The closer the MCC is
to 1, the closer the prediction is to a perfect prediction. to 1, the closer the prediction is to a perfect prediction.

RESULTS AND DISCUSSION RESULTS AND DISCUSSION

Prediction performance of different combinations of input vectors was compared. The Prediction performance of different combinations of input vectors was compared. The
testing results are shown in Table 2. When the first cysteine’s oxidation state is encoded, testing results are shown in Table 2. When the first cysteine’s oxidation state is encoded,
the overall accuracy is 87.4%. Encoding the first two cysteines on the N-terminus the overall accuracy is 87.4%. Encoding the first two cysteines on the N-terminus
increases the accuracy to 94.7%. The encodings with the oxidation states of two cysteines increases the accuracy to 94.7%. The encodings with the oxidation states of two cysteines
on the amino terminus and with the global information on the protein chain yielded the on the amino terminus and with the global information on the protein chain yielded the
best accuracy among all the tests. Accuracy of 94.8% and a MCC value of 0.89 are best accuracy among all the tests. Accuracy of 94.8% and a MCC value of 0.89 are
obtained. This might be explained by the critical biological findings that the terminals of obtained. This might be explained by the critical biological findings that the terminals of
proteins usually carry critical signals for the whole protein chain. proteins usually carry critical signals for the whole protein chain.

Table 1. Prediction performance of SVM based on different combination of input vectors Table 1. Prediction performance of SVM based on different combination of input vectors

SVM vectors in encoding SS-sens. SS-spec. SH-sens. SH-spec. Accu. MCC SVM vectors in encoding SS-sens. SS-spec. SH-sens. SH-spec. Accu. MCC

1 st cys 0.72 0.90 0.96 0.87 0.87 0.72 1 st cys 0.72 0.90 0.96 0.87 0.87 0.72

1,2 cys known 0.90 0.94 0.97 0.95 0.95 0.88 1,2 cys known 0.90 0.94 0.97 0.95 0.95 0.88
st nd st nd
Global info +1 ,2 cys 0.91 0.94 0.97 0.95 0.95 0.89 Global info +1 ,2 cys 0.91 0.94 0.97 0.95 0.95 0.89
st nd st nd
Flanking seq. + 1 , 2 cys 0.90 0.94 0.97 0.94 0.95 0.88 Flanking seq. + 1 , 2 cys 0.90 0.94 0.97 0.94 0.95 0.88
1st ,2nd cys and flanking seq. 1st ,2nd cys and flanking seq.
and global info 0.91 0.94 0.97 0.95 0.95 0.89 and global info 0.91 0.94 0.97 0.95 0.95 0.89

Adding vectors for flanking sequences does not improve the accuracy further. Adding vectors for flanking sequences does not improve the accuracy further.
However, by doing so, the sensitivity of free cysteines and specificity of the bonded However, by doing so, the sensitivity of free cysteines and specificity of the bonded
cysteine were compromised slightly. cysteine were compromised slightly.

4 4
28 28

Specificity is calculated by: Specificity is calculated by:

TN x TN x
Spec = (4) Spec = (4)
TN x FPx TN x FPx

And sensitivity is calculated by: And sensitivity is calculated by:

The Matthews correlation coefficient (MCC) [20] is calculated as: The Matthews correlation coefficient (MCC) [20] is calculated as:

MCC= TPx TN x FPx FN x (6) MCC= TPx TN x FPx FN x (6)

(TPx FN x )(TPx FPx )(TN x FPx )(TN x FN x ) (TPx FN x )(TPx FPx )(TN x FPx )(TN x FN x )

RESULTS AND DISCUSSION RESULTS AND DISCUSSION

Table 1. Prediction performance of SVM based on different combination of input vectors Table 1. Prediction performance of SVM based on different combination of input vectors

SVM vectors in encoding SS-sens. SS-spec. SH-sens. SH-spec. Accu. MCC SVM vectors in encoding SS-sens. SS-spec. SH-sens. SH-spec. Accu. MCC

1 st cys 0.72 0.90 0.96 0.87 0.87 0.72 1 st cys 0.72 0.90 0.96 0.87 0.87 0.72

Downloaded From: http://ebooks.asmedigitalcollection.asme.org/ on 01/28/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

5 5
29 29

Between the 1st and 2nd cysteine on the amino end of the chain, the oxidation state Between the 1st and 2nd cysteine on the amino end of the chain, the oxidation state
of second cysteine is more influential to the oxidation states of other cysteines. of second cysteine is more influential to the oxidation states of other cysteines.
Knowledge of the oxidation state of the second cysteine alone combined with the global Knowledge of the oxidation state of the second cysteine alone combined with the global
information and flanking sequence information give a 91% accuracy, which has already information and flanking sequence information give a 91% accuracy, which has already
outperform the best prediction result in literature [13], as is shown in Table 3. outperform the best prediction result in literature [13], as is shown in Table 3.

Table 3: The relative importance of oxidation states of first cysteine and second cysteines from the Table 3: The relative importance of oxidation states of first cysteine and second cysteines from the
amino side of protein chain. (In each testing, general information about the protein hain such as number of amino side of protein chain. (In each testing, general information about the protein hain such as number of
cysteines and number of amino acids on the chain were also included in the encoding) cysteines and number of amino acids on the chain were also included in the encoding)

SVM vectors used SS-Sens. SS-spec SH-sens. SH-spec Accu. MCC SVM vectors used SS-Sens. SS-spec SH-sens. SH-spec Accu. MCC
1st cys only 0.72 0.9 0.96 0.87 0.87 0.72 1st cys only 0.72 0.9 0.96 0.87 0.87 0.72
2nd cys only 0.78 0.91 0.97 0.9 0.9 0.77 2nd cys only 0.78 0.91 0.97 0.9 0.9 0.77
2nd cys & flanking 0.8 0.91 0.97 0.9 0.9 0.78 2nd cys & flanking 0.8 0.91 0.97 0.9 0.9 0.78
2nd cys & flanking & 2nd cys & flanking &
global 0.81 0.92 0.96 0.9 0.91 0.8 global 0.81 0.92 0.96 0.9 0.91 0.8
1st and 2nd cys & 1st and 2nd cys &
flanking & global 0.91 0.94 0.97 0.95 0.95 0.89 flanking & global 0.91 0.94 0.97 0.95 0.95 0.89

The way how flanking sequences was represented took a role too. As is shown in The way how flanking sequences was represented took a role too. As is shown in
Table 4, when represented by frequency, the accuracy is about 2 % higher when Table 4, when represented by frequency, the accuracy is about 2 % higher when
represented by the order they appear. represented by the order they appear.

Table 4: Influence of how flankcing sequences were represented in the encoding. (Also included in the Table 4: Influence of how flankcing sequences were represented in the encoding. (Also included in the
encoding are 1st and 2nd cysteines’ oxidation state, global information and general information about the chain, encoding are 1st and 2nd cysteines’ oxidation state, global information and general information about the chain,
such as number of cysteines and number of amino acids on the chain) such as number of cysteines and number of amino acids on the chain)

SVM vectors used SS sens. SS spec. SH sens SH spec accu. MCC SVM vectors used SS sens. SS spec. SH sens SH spec accu. MCC
flanking by flanking by
frequency 0.91 0.94 0.97 0.95 0.95 0.89 frequency 0.91 0.94 0.97 0.95 0.95 0.89
flanking by order 0.86 0.94 0.97 0.93 0.93 0.85 flanking by order 0.86 0.94 0.97 0.93 0.93 0.85

CONCLUSIONS CONCLUSIONS
In this research, support vector machine was used for the prediction of bonding states of In this research, support vector machine was used for the prediction of bonding states of
cysteines. We found that the oxidation states of amino terminal cysteines infer the cysteines. We found that the oxidation states of amino terminal cysteines infer the
oxidation states of other cysteines in the same protein chain. By including the oxidations oxidation states of other cysteines in the same protein chain. By including the oxidations
states of up to two cysteines on the amino terminus, accuracy of 95% is achieved. This states of up to two cysteines on the amino terminus, accuracy of 95% is achieved. This
might be explained by biological findings that the terminals of proteins usually carry might be explained by biological findings that the terminals of proteins usually carry
critical signals. Between the 1st and 2nd cysteine on the amino end of the chain, the critical signals. Between the 1st and 2nd cysteine on the amino end of the chain, the
oxidation state of second cysteine is more influential to the oxidation states of other oxidation state of second cysteine is more influential to the oxidation states of other
cysteines. Higher accuracy is obtained when the flanking sequences are represented by cysteines. Higher accuracy is obtained when the flanking sequences are represented by
the frequency of each amino acid. the frequency of each amino acid.

ACKNOWLEDGEMENT ACKNOWLEDGEMENT
This research was supported in part by the National Science Foundation (NSF) under This research was supported in part by the National Science Foundation (NSF) under
Grants CCF-0514750, and CCF-0646102, and a GSU Research Program Enhancement Grants CCF-0514750, and CCF-0646102, and a GSU Research Program Enhancement
Grant. Grant.

5 5
29 29

Downloaded From: http://ebooks.asmedigitalcollection.asme.org/ on 01/28/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

6 6
30 30

REFERENCES REFERENCES

[1] Nagahara, N., Yoshii, T., Abe, Y., Matsumura, T., Thioredoxin-dependent enzymatic [1] Nagahara, N., Yoshii, T., Abe, Y., Matsumura, T., Thioredoxin-dependent enzymatic
activation of mercaptopyruvate sulfurtransferase: An intersubunit disulfide bond serves as a activation of mercaptopyruvate sulfurtransferase: An intersubunit disulfide bond serves as a
redox switch for activation.J Biol Chem. 2006 Nov 27; [Epub ahead of print] redox switch for activation.J Biol Chem. 2006 Nov 27; [Epub ahead of print]
[2] Mannick, JB., Regulation of apoptosis by protein S-nitrosylation. Amino Acids. 2006 Nov [2] Mannick, JB., Regulation of apoptosis by protein S-nitrosylation. Amino Acids. 2006 Nov
30; [Epub ahead of print]. 30; [Epub ahead of print].
[3] Vielle, C., Zeikus, G., Hyperthermophilic enzymes: sources, uses and molecular mechanisms [3] Vielle, C., Zeikus, G., Hyperthermophilic enzymes: sources, uses and molecular mechanisms
for thermostability, Microbiol Mol Biol Rew 2001; 65:1-43 for thermostability, Microbiol Mol Biol Rew 2001; 65:1-43
[4] Hogg, P. J., Disulfide bonds as switches for protein function, Trends Biochem Sci. (2003) [4] Hogg, P. J., Disulfide bonds as switches for protein function, Trends Biochem Sci. (2003)
210-214 210-214
[5] Mitsunori, F., Eiko, K., and Katsuhiko, M., Conserved N-terminal Cysteine Motif Is [5] Mitsunori, F., Eiko, K., and Katsuhiko, M., Conserved N-terminal Cysteine Motif Is
Essential for Homo- and Heterodimer Formation of Synaptotagmins III, V, VI, and X, J Biol Essential for Homo- and Heterodimer Formation of Synaptotagmins III, V, VI, and X, J Biol
Chem, (1999) 31421-31427 Chem, (1999) 31421-31427
[6] Chuang, CC., Chen, CY., Yang, J-M., Lyu, PC., Hwang, JK., Relationship between protein [6] Chuang, CC., Chen, CY., Yang, J-M., Lyu, PC., Hwang, JK., Relationship between protein
structures and disulfide-bonding patterns, Proteins (2003) 1-5 structures and disulfide-bonding patterns, Proteins (2003) 1-5
[7] Muskal, S., Holbrook, S. and Kim, S., Prediction of the disulfide-bonding state of cysteine in [7] Muskal, S., Holbrook, S. and Kim, S., Prediction of the disulfide-bonding state of cysteine in
proteins, Prot. Eng., (1990) 667-672 proteins, Prot. Eng., (1990) 667-672
[8] Fiser, A., Cserzo, M., Tudos, E., Simon, I., Different sequence environments of cysteines and [8] Fiser, A., Cserzo, M., Tudos, E., Simon, I., Different sequence environments of cysteines and
half cystines in proteins. Application to predict disulfide forming residues., FEBS Lett. half cystines in proteins. Application to predict disulfide forming residues., FEBS Lett.
(1992) 117-20 (1992) 117-20
[9] Muccielli-Giorgi, MH., Hazout, S., Tuffery, P.. Predicting the disulfide bonding state of [9] Muccielli-Giorgi, MH., Hazout, S., Tuffery, P.. Predicting the disulfide bonding state of
cysteines using protein descriptors. Proteins, (2002) 243-249 cysteines using protein descriptors. Proteins, (2002) 243-249
[10] Martelli, P.L., Fariselli, P., Malaguti, L. and Casadio, R., Prediction of the disulfide-bonding [10] Martelli, P.L., Fariselli, P., Malaguti, L. and Casadio, R., Prediction of the disulfide-bonding
state of cysteines in proteins at 88% accuracy, Protein Science (2002), 2735-2739 state of cysteines in proteins at 88% accuracy, Protein Science (2002), 2735-2739
[11] Martell, P. L., Fariselli, P., Malaguti, L. and Casadio, R., Prediction of the disulfide bonding [11] Martell, P. L., Fariselli, P., Malaguti, L. and Casadio, R., Prediction of the disulfide bonding
state of cysteines in proteins with hidden neural networks, Protein Engineering, (2002) 951- state of cysteines in proteins with hidden neural networks, Protein Engineering, (2002) 951-
953 953
[12] Frasconi, P., Passerini, A., Vullo, A. , A two-stage SVM architecture for predicting the [12] Frasconi, P., Passerini, A., Vullo, A. , A two-stage SVM architecture for predicting the
disulfide bonding state of cysteines, Neural Networks for Signal Processing, 2002. disulfide bonding state of cysteines, Neural Networks for Signal Processing, 2002.
Proceedings of the 2002 12th IEEE Workshop on, 25-34 Proceedings of the 2002 12th IEEE Workshop on, 25-34
[13] Chen, YC., Lin, YS., Lin, CJ. and Hwang, JK., Prediction of the bonding states of cysteines [13] Chen, YC., Lin, YS., Lin, CJ. and Hwang, JK., Prediction of the bonding states of cysteines
using the support vector machines based on multiple feature vectors and cysteine state using the support vector machines based on multiple feature vectors and cysteine state
sequences, Proteins: structure, function and bioinformatics (2004) 1036-1042 sequences, Proteins: structure, function and bioinformatics (2004) 1036-1042
[14] Ceroni, A., Passerini, A., Vullo, A., and Frasconi, P., DISULFIND: a disulfide [14] Ceroni, A., Passerini, A., Vullo, A., and Frasconi, P., DISULFIND: a disulfide
bonding state and cysteine connectivity prediction server, nucleic acids research, bonding state and cysteine connectivity prediction server, nucleic acids research,
(2006) W177-W181 (2006) W177-W181
[15] Cristianini, N.., and Shawe-Taylor, J., An Introduction to Support Vector Machines. [15] Cristianini, N.., and Shawe-Taylor, J., An Introduction to Support Vector Machines.
Cambridge University Press (2000) Cambridge University Press (2000)
[16] Vapnik, V. and Cortes, C., Support vector networks. Machine Learning (1995) 273-293 [16] Vapnik, V. and Cortes, C., Support vector networks. Machine Learning (1995) 273-293
[17] Joachims T., Making large-Scale SVM Learning Practical. Advances in Kernel Methods - [17] Joachims T., Making large-Scale SVM Learning Practical. Advances in Kernel Methods -
Support Vector Learning, B. Schölkopf and C. Burges and A. Smola (ed.), MIT-Press, (1999) Support Vector Learning, B. Schölkopf and C. Burges and A. Smola (ed.), MIT-Press, (1999)
[18] Meiler, J., Muller, M., Zeidler, A., Schmaschke, F. Generation and evaluation of dimension- [18] Meiler, J., Muller, M., Zeidler, A., Schmaschke, F. Generation and evaluation of dimension-
reduced amino acid parameter representations by artificial neural networks, J Mol Model reduced amino acid parameter representations by artificial neural networks, J Mol Model
(2001) 360:369 (2001) 360:369
[19] Vapnik, V., The nature of statistical learning theory. New York: Springer; (1995) [19] Vapnik, V., The nature of statistical learning theory. New York: Springer; (1995)
[20] Matthews, B. W., Comparison of the predicted and observed secondary structure of [20] Matthews, B. W., Comparison of the predicted and observed secondary structure of
T4 phage lysozyme. Biochim Biophys Acta (1975) 442-451 T4 phage lysozyme. Biochim Biophys Acta (1975) 442-451

6 6
30 30

REFERENCES REFERENCES

Downloaded From: http://ebooks.asmedigitalcollection.asme.org/ on 01/28/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

31 31

FUZZY NEURAL NETWORKS FOR DIAGNOSIS OF MALIGNANT FUZZY NEURAL NETWORKS FOR DIAGNOSIS OF MALIGNANT
MESOTHELIOMA MESOTHELIOMA

ARUN KULKARNI MADHUKAR BANDI ARUN KULKARNI MADHUKAR BANDI

Computer Science Department Computer Science Department Computer Science Department Computer Science Department
The University of Texas at Tyler The University of Texas at Tyler, The University of Texas at Tyler The University of Texas at Tyler,
Tyler, TX 75799 Tyler, TX 75799 Tyler, TX 75799 Tyler, TX 75799
[email protected] [email protected] [email protected] [email protected]

ABSTRACT ABSTRACT
Many computer based diagnostic systems are being used in practice in areas Many computer based diagnostic systems are being used in practice in areas
such as the mammography, analysis of magnetic resonance imaging (MRI), and such as the mammography, analysis of magnetic resonance imaging (MRI), and
computer tomography. However, the diagnosis of pathological images is still computer tomography. However, the diagnosis of pathological images is still
more of an art, and it is based on the pathologist’s knowledge, experience, and more of an art, and it is based on the pathologist’s knowledge, experience, and
intuition. In pathology, suspect tissue samples are removed from the host and intuition. In pathology, suspect tissue samples are removed from the host and
are stained before they are observed under the microscope. In this paper, we are stained before they are observed under the microscope. In this paper, we
propose a computer based diagnostic system to analyze images of stained propose a computer based diagnostic system to analyze images of stained
samples. The proposed system tries to mimic the human vision and decision samples. The proposed system tries to mimic the human vision and decision
making process. During preprocessing, pixels are mapped from the red-blue- making process. During preprocessing, pixels are mapped from the red-blue-
green (RGB) color space to the hue-saturation-intensity (HIS) feature space. green (RGB) color space to the hue-saturation-intensity (HIS) feature space.
The feature extraction stage deals with obtaining color features from the The feature extraction stage deals with obtaining color features from the
segmented images. During the decision making, the images are separated in segmented images. During the decision making, the images are separated in
two classes positive and negatives. two classes positive and negatives.

BACKGROUND BACKGROUND
Computer based diagnostic systems are commonly used by radiologists in areas such Computer based diagnostic systems are commonly used by radiologists in areas such
as the mammography, analysis of magnetic resonance images (MRI), and computer as the mammography, analysis of magnetic resonance images (MRI), and computer
tomography. There are many advantages of using a computer diagnostic system. First, it tomography. There are many advantages of using a computer diagnostic system. First, it
yields consistent, objective, and repeatable results that are free from intra and inter yields consistent, objective, and repeatable results that are free from intra and inter
observer variations. Secondly, unlike the human eye, which can differentiate only limited observer variations. Secondly, unlike the human eye, which can differentiate only limited
number of color intensity strengths, a computer diagnostic system can analyze the image number of color intensity strengths, a computer diagnostic system can analyze the image
pixel-by-pixel with a large number of color intensity values. Thirdly, a computer pixel-by-pixel with a large number of color intensity values. Thirdly, a computer
diagnostic system can handle a large amount of data with out fatigue and the disdain for diagnostic system can handle a large amount of data with out fatigue and the disdain for
repetitiveness. Until recently, most computer diagnostic systems have been designed for repetitiveness. Until recently, most computer diagnostic systems have been designed for
radiology to assist radiologists in decision-making. Computer diagnostic systems have radiology to assist radiologists in decision-making. Computer diagnostic systems have
not made their way to diagnosis of pathological images. The diagnosis of pathological not made their way to diagnosis of pathological images. The diagnosis of pathological
images is based on the pathologist’s knowledge, experience, and intuition. One of the images is based on the pathologist’s knowledge, experience, and intuition. One of the
significant differences between images in radiology and images in pathology is that most significant differences between images in radiology and images in pathology is that most
images in radiology are black-and-white and are obtained by some scanning mechanism. images in radiology are black-and-white and are obtained by some scanning mechanism.
However, most pathological images are color images and are obtained by observing However, most pathological images are color images and are obtained by observing
stained tissue samples under the microscope. Tissue samples are removed from the host stained tissue samples under the microscope. Tissue samples are removed from the host
and are stained before they are observed under the microscope. The process of staining and are stained before they are observed under the microscope. The process of staining
samples modifies the color of the tissue sample. Images of stained tissue samples are samples modifies the color of the tissue sample. Images of stained tissue samples are

31 31

FUZZY NEURAL NETWORKS FOR DIAGNOSIS OF MALIGNANT FUZZY NEURAL NETWORKS FOR DIAGNOSIS OF MALIGNANT
MESOTHELIOMA MESOTHELIOMA

ARUN KULKARNI MADHUKAR BANDI ARUN KULKARNI MADHUKAR BANDI

Downloaded From: http://ebooks.asmedigitalcollection.asme.org/ on 01/28/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

32 32

used for diagnosis. In this paper, we considere a problem of designing a computer - used for diagnosis. In this paper, we considere a problem of designing a computer -
based system to analyze pathological images for diagnosis of malignant mesothelioma based system to analyze pathological images for diagnosis of malignant mesothelioma
(MM). Malignant mesothelioma is an aggressive neoplasm for which current therapy is (MM). Malignant mesothelioma is an aggressive neoplasm for which current therapy is
unsatisfactory. This tumor causes up to 4000 deaths per year in the United States and it is unsatisfactory. This tumor causes up to 4000 deaths per year in the United States and it is
a world wide health problem associated with asbestos exposure (Idell et al., 1995). In our a world wide health problem associated with asbestos exposure (Idell et al., 1995). In our
approach, we try to mimics pathologists’ vision and decision process. The key factor that approach, we try to mimics pathologists’ vision and decision process. The key factor that
pathologists use in decision making is the difference between the color of cell images and pathologists use in decision making is the difference between the color of cell images and
the background. In our implementation, we use color clustering for segmentation. Jiang, the background. In our implementation, we use color clustering for segmentation. Jiang,
et al. (2006) have used color clustering for segmentation. We first segment images and et al. (2006) have used color clustering for segmentation. We first segment images and
then extract color features that distinguish cell areas and the background. Finally, we use then extract color features that distinguish cell areas and the background. Finally, we use
a fuzzy neural network for decision making. We train the network using training a fuzzy neural network for decision making. We train the network using training
samples. We have used twenty microscopic images of stained tissue samples obtained samples. We have used twenty microscopic images of stained tissue samples obtained
from the pathological laboratory at The University of Texas Health Center at Tyler. from the pathological laboratory at The University of Texas Health Center at Tyler.

METHODOLOGY METHODOLOGY
The proposed system consists of three stages the preprocessing, feature extraction, The proposed system consists of three stages the preprocessing, feature extraction,
and decision making. During the preprocessing stage we map pixels from the red, blue, and decision making. During the preprocessing stage we map pixels from the red, blue,
green (RBG) color space to hue, saturation, intensity (HIS) color space. In a color image, green (RBG) color space to hue, saturation, intensity (HIS) color space. In a color image,
each pixel is characterized by a great number of combinations of RGB chromatic each pixel is characterized by a great number of combinations of RGB chromatic
components. However, more complicated segmentation techniques are required to deal components. However, more complicated segmentation techniques are required to deal
with rich chromatic information in the segmentation of color images. A number of color with rich chromatic information in the segmentation of color images. A number of color
spaces have been proposed for segmentation of color images. The HIS color space is spaces have been proposed for segmentation of color images. The HIS color space is
more suitable for segmentation because the RGB color space is far from exhibiting the more suitable for segmentation because the RGB color space is far from exhibiting the
perceptual uniformity and it does not model the way human perceive the color (Dong and perceptual uniformity and it does not model the way human perceive the color (Dong and
Xie, 2005). In feature extraction stage, we segment the color image using unsupervised Xie, 2005). In feature extraction stage, we segment the color image using unsupervised
clustering. We use a fuzzy neural network with competitive learning algorithm for clustering. We use a fuzzy neural network with competitive learning algorithm for
segmentation, and a perceptron model for decision making.. segmentation, and a perceptron model for decision making..

RGB to HIS Mapping RGB to HIS Mapping

The RGB to HIS color space transformation is described by following equations The RGB to HIS color space transformation is described by following equations
(Gonzalez and Woods, 2002). (Gonzalez and Woods, 2002).

The intensity and saturation components are given by The intensity and saturation components are given by
1 1
I RG B (1) I RG B (1)
3 3

The saturation component S is given by The saturation component S is given by

3 3
S 1 > min R, G , B @ (2) S 1 > min R, G , B @ (2)
RG B RG B

The hue component H is given by The hue component H is given by

T if B d G ½ T if B d G ½
H ® ¾ (3) H ® ¾ (3)
¯ 2S -T if B > G¿ ¯ 2S -T if B > G ¿

32 32

RGB to HIS Mapping RGB to HIS Mapping

The intensity and saturation components are given by The intensity and saturation components are given by
1 1
I RG B (1) I RG B (1)
3 3

The saturation component S is given by The saturation component S is given by

3 3
S 1 > min R, G , B @ (2) S 1 > min R, G , B @ (2)
RG B RG B

The hue component H is given by The hue component H is given by

T if B d G ½ T if B d G ½
H ® ¾ (3) H ® ¾ (3)
¯ 2S -T if B > G¿ ¯ 2S -T if B > G ¿

Downloaded From: http://ebooks.asmedigitalcollection.asme.org/ on 01/28/2016 Terms of Use: http://www.asme.org/about-asme/terms-of-use

Exploring the Variety of Random
Documents with Different Content
back
back
back
back
back
back
back
back
back
back
back
back
back
back
back
Welcome to Our Bookstore - The Ultimate Destination for Book Lovers
Are you passionate about books and eager to explore new worlds of
knowledge? At our website, we offer a vast collection of books that
cater to every interest and age group. From classic literature to
specialized publications, self-help books, and children’s stories, we
have it all! Each book is a gateway to new adventures, helping you
expand your knowledge and nourish your soul
Experience Convenient and Enjoyable Book Shopping Our website is more
than just an online bookstore—it’s a bridge connecting readers to the
timeless values of culture and wisdom. With a sleek and user-friendly
interface and a smart search system, you can find your favorite books
quickly and easily. Enjoy special promotions, fast home delivery, and
a seamless shopping experience that saves you time and enhances your
love for reading.
Let us accompany you on the journey of exploring knowledge and
personal growth!

ebookgate.com

Handbook On Computational Intelligence (In 2 Volumes) PDF
100% (3)
Handbook On Computational Intelligence (In 2 Volumes) PDF
957 pages
Fuzzy Systems, Neural Networks and Neuro-Fuzzy Systems A Vision On Their Hardware Implementation and Platforms Over Two Decades
No ratings yet
Fuzzy Systems, Neural Networks and Neuro-Fuzzy Systems A Vision On Their Hardware Implementation and Platforms Over Two Decades
49 pages
Ali Ahmadian (Editor), Soheil Salahshour (Editor) - Soft Computing Approach For Mathematical Modeling of Engineering Problems-CRC Press (2021)
100% (2)
Ali Ahmadian (Editor), Soheil Salahshour (Editor) - Soft Computing Approach For Mathematical Modeling of Engineering Problems-CRC Press (2021)
267 pages
Fuzzy Logic Neural Network Genetic Algorithms PDF
100% (1)
Fuzzy Logic Neural Network Genetic Algorithms PDF
381 pages
Design of Intelligent Systems Based On Fuzzy Logic, Neural Networks and Nature-Inspired Optimization
No ratings yet
Design of Intelligent Systems Based On Fuzzy Logic, Neural Networks and Nature-Inspired Optimization
612 pages
Artificial Intelligence & Machine Learning Notes
No ratings yet
Artificial Intelligence & Machine Learning Notes
177 pages
Computational Intelligence in Architecting Engineering Systems 1st Edition C. H. Dagli Download
No ratings yet
Computational Intelligence in Architecting Engineering Systems 1st Edition C. H. Dagli Download
55 pages
Soft Computing
No ratings yet
Soft Computing
6 pages
Intelligent Information Processing With Matlab - Xiu Zhang
No ratings yet
Intelligent Information Processing With Matlab - Xiu Zhang
347 pages
Artificial Neural Networks
No ratings yet
Artificial Neural Networks
367 pages
A C I S V 6 S E: J L, J M: Tlantis Omputational Ntelligence Ystems Olume
No ratings yet
A C I S V 6 S E: J L, J M: Tlantis Omputational Ntelligence Ystems Olume
677 pages
R20CSE3122 Artificial Intelligence
No ratings yet
R20CSE3122 Artificial Intelligence
140 pages
Enhancing Tech Theory
From Everand
Enhancing Tech Theory
T. T. Samuels
No ratings yet
(Artificial Intelligence Notes
No ratings yet
(Artificial Intelligence Notes
140 pages
A8702 Artificial-Intelligence-ALL Notes
No ratings yet
A8702 Artificial-Intelligence-ALL Notes
113 pages
Artificial Intelligence and Machine
100% (1)
Artificial Intelligence and Machine
328 pages
Fuzzy Optimization Techniques in The Areas of Science and Management 1st Edition Santosh Kumar Das All Chapter Instant Download
No ratings yet
Fuzzy Optimization Techniques in The Areas of Science and Management 1st Edition Santosh Kumar Das All Chapter Instant Download
49 pages
Mathematical Problems in Engineering - 2012 - Lu - Artificial Intelligence in Civil Engineering
No ratings yet
Mathematical Problems in Engineering - 2012 - Lu - Artificial Intelligence in Civil Engineering
22 pages
Jamshid Ghaboussi - Soft Computing in Engineering-CRC Press (2018)
No ratings yet
Jamshid Ghaboussi - Soft Computing in Engineering-CRC Press (2018)
221 pages
ASCE Newsletter
No ratings yet
ASCE Newsletter
9 pages
Artificial Intelligence Robotics MCQ: Ads by
No ratings yet
Artificial Intelligence Robotics MCQ: Ads by
68 pages
(R17A1204) Artificial Intelligence
No ratings yet
(R17A1204) Artificial Intelligence
257 pages
Ai Unit 1 Notes
No ratings yet
Ai Unit 1 Notes
21 pages
Basic Artificial Intelligence Concepts in Civil Engineering
No ratings yet
Basic Artificial Intelligence Concepts in Civil Engineering
30 pages
Ai Module - 1
No ratings yet
Ai Module - 1
33 pages
SC Unit-1
No ratings yet
SC Unit-1
58 pages
Tech Seminar
No ratings yet
Tech Seminar
20 pages
Rizal 19th Century New
No ratings yet
Rizal 19th Century New
38 pages
Fuzzy Logic Neural Network Genetic Algorithms
No ratings yet
Fuzzy Logic Neural Network Genetic Algorithms
381 pages
Page 7
No ratings yet
Page 7
26 pages
Foundations of Neural Networks Fuzzy Sys PDF
No ratings yet
Foundations of Neural Networks Fuzzy Sys PDF
1 page
Long-Range Planning and Strategic Management
No ratings yet
Long-Range Planning and Strategic Management
9 pages
Gainsforth 1945
100% (1)
Gainsforth 1945
12 pages
BIMBOT-Artificial Intelligence Applied To BIM Desi
No ratings yet
BIMBOT-Artificial Intelligence Applied To BIM Desi
16 pages
Presentasi Perkuliahan 1
No ratings yet
Presentasi Perkuliahan 1
26 pages
Intelligent Controllers
No ratings yet
Intelligent Controllers
33 pages
Soft Computing
No ratings yet
Soft Computing
41 pages
Mathematical Problems in Engineering - 2012 - Lu - Artificial Intelligence in Civil Engineering
No ratings yet
Mathematical Problems in Engineering - 2012 - Lu - Artificial Intelligence in Civil Engineering
22 pages
Artificial Neural Network
No ratings yet
Artificial Neural Network
11 pages
Wseas TS2004
No ratings yet
Wseas TS2004
10 pages
Artificial Intelligence in Civil Engineering: Presented by Aseena.L Roll No:20 S7 Ce Mcet
No ratings yet
Artificial Intelligence in Civil Engineering: Presented by Aseena.L Roll No:20 S7 Ce Mcet
29 pages
Unit I
No ratings yet
Unit I
114 pages
Psychology Chapter 7: Human Memory
No ratings yet
Psychology Chapter 7: Human Memory
26 pages
Digital Logic Design 6 Counters
No ratings yet
Digital Logic Design 6 Counters
37 pages
Rudas 2003
No ratings yet
Rudas 2003
8 pages
Collective Artificial Intelligence
From Everand
Collective Artificial Intelligence
Diego Sonvico
No ratings yet
Research Article: Urban Planning and Design Layout Generation Based On Artificial Intelligence
No ratings yet
Research Article: Urban Planning and Design Layout Generation Based On Artificial Intelligence
10 pages
Artificial Intelligence Applications in Civil Construction Architectural Engineering Education
No ratings yet
Artificial Intelligence Applications in Civil Construction Architectural Engineering Education
10 pages
Jio Final Report
No ratings yet
Jio Final Report
55 pages
Eie 557
No ratings yet
Eie 557
3 pages
Chapter 1 Overview of AI
No ratings yet
Chapter 1 Overview of AI
45 pages
Bad Debt Policy
No ratings yet
Bad Debt Policy
5 pages
Current Trends
No ratings yet
Current Trends
15 pages
List of Documents To Be Attached With The Application Form For Registration As Professional Engineer (Pe) (Through Epe)
100% (1)
List of Documents To Be Attached With The Application Form For Registration As Professional Engineer (Pe) (Through Epe)
6 pages
Labeeb Khan Resume
No ratings yet
Labeeb Khan Resume
2 pages
Publicaciones Relacionadas IA
No ratings yet
Publicaciones Relacionadas IA
40 pages
Application in Mechanical Engg - Paper 1
No ratings yet
Application in Mechanical Engg - Paper 1
7 pages
Soft Computing Introduction
No ratings yet
Soft Computing Introduction
29 pages
Soft Computing: Presented BY: Ganesh Paul TT - IT
No ratings yet
Soft Computing: Presented BY: Ganesh Paul TT - IT
20 pages
Sample Formatted Document - Research Report
No ratings yet
Sample Formatted Document - Research Report
5 pages
About The CD-ROM: Pathfinding Algorithm
No ratings yet
About The CD-ROM: Pathfinding Algorithm
6 pages
Intelligent Systems
No ratings yet
Intelligent Systems
3 pages
cpphtp10 07
No ratings yet
cpphtp10 07
111 pages
Menstural Cycle DISORDERS
No ratings yet
Menstural Cycle DISORDERS
28 pages
Decluttering For Dummies, Portable Edition Jane Stoller - Read The Ebook Online or Download It For A Complete Experience
100% (1)
Decluttering For Dummies, Portable Edition Jane Stoller - Read The Ebook Online or Download It For A Complete Experience
54 pages
Assignment 5 Ai
No ratings yet
Assignment 5 Ai
3 pages
Artificial Intelligence Bayesian Networks
No ratings yet
Artificial Intelligence Bayesian Networks
6 pages
Fuzzy Systems in Bio-Inspired Computing - State-of-the-Art Literature Review
No ratings yet
Fuzzy Systems in Bio-Inspired Computing - State-of-the-Art Literature Review
1 page
DainikBhaskar - Innovating Its Way To Success
No ratings yet
DainikBhaskar - Innovating Its Way To Success
7 pages
Gujarat Technological University: W.E.F. AY 2018-19
No ratings yet
Gujarat Technological University: W.E.F. AY 2018-19
3 pages
Artificial Intelligence in Power Station
No ratings yet
Artificial Intelligence in Power Station
2 pages
REPSOL GXR 5 - 1 v1
No ratings yet
REPSOL GXR 5 - 1 v1
1 page
Chapter 1. M.F
No ratings yet
Chapter 1. M.F
11 pages
QKD QKTD
No ratings yet
QKD QKTD
6 pages
Pto Muncie FORD-10 PDF
No ratings yet
Pto Muncie FORD-10 PDF
2 pages
101 Newsletter Content Ideas - Pages
No ratings yet
101 Newsletter Content Ideas - Pages
13 pages
Contents Artificial Intelligence
No ratings yet
Contents Artificial Intelligence
3 pages
Performance of Sound Insulation of AAC in Massive Buildings-Experience With EN 12354-1
No ratings yet
Performance of Sound Insulation of AAC in Massive Buildings-Experience With EN 12354-1
6 pages
Nicolaus Copernicus
No ratings yet
Nicolaus Copernicus
11 pages
7A Concept Map Cells
100% (1)
7A Concept Map Cells
4 pages
Bphet 141 1
No ratings yet
Bphet 141 1
12 pages
Speeding Up
No ratings yet
Speeding Up
13 pages
Carbohydrates Pcog
No ratings yet
Carbohydrates Pcog
8 pages
Strategic Management (SM) : Saurav Banerjee 1
No ratings yet
Strategic Management (SM) : Saurav Banerjee 1
11 pages
AKI 2 Primary Care Bundle Oxfordshire V 1.1
No ratings yet
AKI 2 Primary Care Bundle Oxfordshire V 1.1
1 page
Symbolic Artificial Intelligence: Fundamentals and Applications
From Everand
Symbolic Artificial Intelligence: Fundamentals and Applications
Fouad Sabry
No ratings yet
CE OOO BOQ Solar PV System Contractor XXX
No ratings yet
CE OOO BOQ Solar PV System Contractor XXX
1 page
Vaad Vivad - Round 2: Questions
No ratings yet
Vaad Vivad - Round 2: Questions
2 pages
Poster Presentation-Assessment Rubric: Group: Class
No ratings yet
Poster Presentation-Assessment Rubric: Group: Class
2 pages
Hybrid Intelligent System: Fundamentals and Applications
From Everand
Hybrid Intelligent System: Fundamentals and Applications
Fouad Sabry
No ratings yet