Novel Hybrid Neural Agents for Immunoinformatics

1
Novel Hybrid Neural Agents for Immunoinformatics


KHRIZEL B. SOLANO Union County College/
New Jersey Institute of Technology
kbs4_at_njit.edu
TOLJA DJEKOVIC Oakland University
tdjekovi_at_oakland.edu
DR. MOHAMED ZOHDY Oakland University
zohdyma_at_oakland.edu
UnCoRe Program, Oakland University, Summer 2005
2
Motivation
Biological systems may be used as an analogy to
create novel computer paradigms.
   
Nervous System
Immune System
Artificial Neural
Networks
Artificial Immune System
As such, computing may also be used as an
effective tool to study biological systems.
3
Immune System
4
Immune System Concept Map
5
Immunoinformatics

• What is it?
• Aims to explore computational methods for the
  interaction of the immune system components
• Uses Biology for deeper insight of living
  organisms and Computer Science to store and
  analyze enormous amount of data
• Why Immunoinformatics?
• Increased advances in both biological and
  computer research!

6
Artificial Neural Networks (Agents)

• What is it?
• A computer paradigm, inspired by the way human
  nervous system handles information
• Why use it?
• Adaptive learning
• Self-organizing
• Operating in real time
• Fault Tolerance via Redundant Information Coding

7
Artificial Neural Networks (Agents)

• Where is it applied?
• Economy
• Military
• Industry
• Computing
• Biology
• Why neural agents?
• In our application to the immune system, we may
  view a set of neural networks (agents) to
  correspond to the set of immune system components

8
PIE Paradigm
In addition, our neural agents utilize so-called
PIE Paradigm which stands for

• Polymorphism
• Combination types of several neural networks
  (Feed-forward, Hopfield and Elman neural agents)
• Inheritance
• Simple inherit training patterns
• Advanced inherit training sub-functions
• Encapsulation
• Group several sub-functions into one

9
Types of ANN Used Feed-forward Network

• Architecture Input nodes receive information and
  pass it on to middle nodes then along to output
  nodes
• Mainly used for function approximation
• For our immune system research, they are employed
  to mathematically represent
• Immunity initiation
• Self or non-self discrimination
• Immunity recognition
• High level Macrophages
• Medium level B-cell receptors
• Low level T-cell receptors

10
Types of ANN Used Hopfield Network

• Architecture Fully interconnected feedback
  neural network trained to converge to desired
  equilibrium quantized patterns
• For our immune system research, they are employed
  to mathematically represent
• Evolution of B-cells/T-cells
• Cell Selection
• Cell Cloning
• Cell Mutation

11
Types of ANN Used Elman Network

• Architecture two-layer network with feedback and
  time-delay from the output layer to the input
  layer
• Used for temporal or spatial calculations
• For our immune system research, they are employed
  to mathematically represent
• Immune adaptive response
• Cell interaction functional representation
• Immune system differential equation solution

12
Goals

• To configure and program our PIE paradigm neural
  agents such that they have the capability of
  simulating and predicting stable, unstable and
  oscillatory immune system temporal and spatial
  responses
• To optimize the accuracy of training and learning
  the respective neural agents through the PIE
  Paradigm
• To use concise functional block representation of
  the immune system
• To effectively use MATLABs Neural Network
  Toolbox
• To effectively use MATLABs Bioinformatics
  Toolbox

13
Functional Blocks
To summarize the immune system concepts, we
adopted the following set of functional blocks
14
Preliminary Results Feed-forward Network
1 Initiation Comparing
Self and Non-self
The Self/Non-self Model
2 Recognition
Comparing Self and Non-self
3 Recognition
Comparing Self and Non-self
15
Preliminary Results Feed-forward Network
Feed-forward network was used to train and
simulate 1, 2 and 3
Training of 1
Training of 2
Training of 3
Output of 1
Output of 2
Output of 3
16
Preliminary Results Hopfield Network
Cell Selection Model
Hopfield network was used
17
Differential Equation Models 2nd Order
The model of B lymphocytes with virus present



where k virus killing
factor

r
virus growth factor

K
B-cells cloning rate


T number of T-cells





µ B-cell death rate


a, b interaction factors

? B-cell growth rate

dB/dt the rate of
B-cells change over time

dV/dt the rate of Virus over time

18
Differential Equation Models
Interaction of B and T lymphocytes with virus
present
19
Differential Equation Models
Interaction of B-cells to a specific virus, B0
Activated B-cells, B1
Cloning B-cells, Bi
Neutralizing IgM antibodies, M
Virus clearance, V,
20

Preliminary Results Elman Network
Training and output for the function x(k1)
(1-T)x(k), where T is the sampling time interval
and has values of T0.5, T0.05 and T0.005
respectively
Training
Output
21
Preliminary Results Elman Network
Training
Output
Training
Output
22
Plans for the Second Half

• Case studies of immune system diseases VSV,
  SIV, SARS and West Nile
• Use Bioinformatics Toolbox for analysis of
  realistic data
• Relationship of Neural Networks to Mathematical
  modeling

23
Acknowledgement

• Family and friends for their love and support
  
• Dr. Zohdy for his help and guidance
• Dr. Sethi, Dr. Mili, Dr. Elhajj, and Dr. Kim
• All UnCoRe Participants
• Aiyesha Ma
• Ann Leone
• Sarah Zohdy

24
References
1 R. M. Anderson and R. M. May, Infectious
Diseases of Humans Dynamics and Control. London
Oxford University Press, 1992. 2 D. Dasgupta,
Artificial neural networks and artificial immune
systems Similarities and differences, in
Proceedings of IEEE System, Man, and Cybernetics
Conference (SMC 97), 1997, pp. 873-878. 3 L.
N. de Castro, Comparing immune and neural
networks, in Proceedings of the VII Brazillian
Symposium on Neural Networks (SBRN 02), 2002.
4 S. Forrest, S. A. Hofmeyr and A. Somayaji,
Computer immunology, Communications of the ACM,
vol. 40, no. 10, October 1997. 5 G. A. Funk, A.
D. Barbour, H. Hengartner and U. Kalinke,
Mathematical model of a virus-neutralizing
immunoglobulin response, Journal of Theoretical
Biology, vol. 195, pp. 41-52, Nov. 1998. 6 A.
J. Graaf and A. P. Engelbrecht, Using a
threshold function to determine the status of
lymphocytes in the artificial immune system, in
Proceedings of SAICSIT, 2003, pp. 268-274. 7
C. A. Janeway Fr., P. Travers, M. Walport and J.
D. Capra, Immunobiology The Immune System in
Health and Diseases. New York Elsevier Science
Ltd/Garland Publishing, 1999. 8 S. Pramanik,
R. Kozma and D. Dasgupta, Dynamical
neuro-representation of an immune system model
and its application for data classification, in
Proceedings of IJCNN, WCCI, 2002, pp.
130-135. Sources for pictures and
animations http//www.marymount.k12.ny.us/marynet
/stwbwk03/03bio/Immune/ http//www.bestanimations.
com/ http//www.infections.bayer.com/en/bacteria/i
mmunesystem/ http//www.rcai.riken.go.jp/eng/group
/host/ http//www.artie.com/ http//www.doc.ic.ac.
uk/nd/surprise_96/journal/vol4/cs11/report.html h
ttp//www.gifanimations.com/ http//www.millan.net
/anims/giffar.html http//www.animationlibrary.com
/ http//www.uselessgraphics.com/
25
The Story of a Mathematical Biologist
A mathematical biologist spends his vacation
hiking in the Scottish highlands. One day, he
encounters a shepherd with a large herd of sheep.
One of these cuddly, woolly animals would make a
great pet, he thinks... How much for one of your
sheep?" he asks the shepherd. "They aren't for
sale", the shepherd replies. The math biologist
ponders for a moment and then says "I will give
you the precise number of sheep in your herd
without counting. If I'm right, don't you think
that I deserve one of them as a reward? The
shepherd nods. The math biologist says "387".
The shepherd is silent for a while and then says
"You're right. I hate to lose any of my sheep,
but I promised One of them is yours. Have your
pick!" The math biologist grabs one of the
animals, puts it on his shoulders, and is about
to march on, when the shepherd says "Wait! I
will tell you what your profession is, and if I'm
right I'll get the animal back. "That's fair
enough." "You must be a mathematical biologist.
The man is stunned. "You're right. But how could
you know?" "That's easy You gave me the precise
number of sheep without counting - and then you
picked my dog..."