an introduction to the artificial immune systems -...
TRANSCRIPT
An Introduction to theArtificial Immune Systems
ICANNGA 2001Prague, 22-25th April, 2001
Leandro Nunes de [email protected]
http://www.dca.fee.unicamp.br/~lnunesState University of Campinas - UNICAMP/Brazil
Financial Support: FAPESP - 98/11333-9
ICANNGA 2001 - An Introduction to the Artificial Immune Systems 2
• Part I– Introduction to the Immune System
• Part II– Artificial Immune Systems (AIS)
• Part III– Examples of AIS and Applications
– Discussion and Future Trends
Presentation Topics
ICANNGA 2001 - An Introduction to the Artificial Immune Systems 3
— Part I —
• Introduction to the Immune System– Fundamentals and Main Components
– Anatomy
– Innate Immune System
– Adaptive Immune System
– Pattern Recognition in the Immune System
– A General Overview of the Immune Defenses
– Clonal Selection and Affinity Maturation
– Self/Nonself Discrimination
– Immune Network Theory
ICANNGA 2001 - An Introduction to the Artificial Immune Systems 4
The Immune System (I)
• Fundamentals:– Immunology is the study of the defense mechanisms
that confer resistance against diseases (Klein, 1990)
– The immune system (IS) is the one responsible toprotect us against the attack from externalmicroorganisms (Tizard, 1995)
– Several defense mechanisms in different levels; someare redundant
– The IS presents learning and memory
– Microorganisms that might cause diseases (pathogen):viruses, funguses, bacteria and parasites
– Antigen: any molecule that can stimulate the IS
ICANNGA 2001 - An Introduction to the Artificial Immune Systems 5
• Innate immune system:– immediately available for combat
• Adaptive immune system:– antibody (Ab) production specific to a determined
infectious agent
The Immune System (II)
G ran u locytes M acrop h ag es
In n ate
B -ce lls T-ce lls
L ym p h ocytes
A d ap ta tive
Im m u n ity
ICANNGA 2001 - An Introduction to the Artificial Immune Systems 6
• AnatomyThe Immune System (III)
Lymphatic vessels
Lymph nodes
Thymus
Spleen
Tonsils andadenoids
Bone marrow
Appendix
Peyer’s patches
Primary lymphoidorgans
Secondary lymphoidorgans
ICANNGA 2001 - An Introduction to the Artificial Immune Systems 7
• All living beings present a type of defensemechanism
• Innate Immune System– first line of defense
– controls bacterial infections
– regulates the adaptive immunity
– important for self/nonself discrimination
– composed mainly of phagocytes and the complementsystem
The Immune System (IV)
ICANNGA 2001 - An Introduction to the Artificial Immune Systems 8
• The Adaptive Immune System– the vertebrates have an anticipatory immune system
– the lymphocytes carry antigen receptors on theirsurfaces. These receptors are specific to a given antigen
– Clonal selection: B-cells that recognize antigens arestimulated, proliferate (clone) and differentiate intomemory and plasma cells
– confer resistance against future infections
– is capable of fine-tuning the cell receptors of theselected cells to the selective antigens
The Immune System (V)
ICANNGA 2001 - An Introduction to the Artificial Immune Systems 9
• Pattern Recognition: B-cell
The Immune System (VI)
B-cell
BCR or Antibody
Epitopes
B-cell Receptors (Ab)
Antigen
• Pattern Recognition: B-cell
ICANNGA 2001 - An Introduction to the Artificial Immune Systems 10
• Pattern Recognition: T-cell
The Immune System (VII)
T-cell
TCR
APC
MHC-II Protein Pathogen
Peptide
TH cell
MHC/peptide complex
ICANNGA 2001 - An Introduction to the Artificial Immune Systems 11
The Immune System (VIII)
APC
MHC protein Pathogen
Peptide
T cell
Activated T cell
B cell
Lymphokines
Activated B cell(Plasma cell)
( I )
( II )
( III )
( IV )
( V )
( VI )
after Nosssal, 1993
ICANNGA 2001 - An Introduction to the Artificial Immune Systems 12
• Antibody Synthesis:
The Immune System (IX)
... ... ... V V
V library
D D
D library
J J
J library
Gene rearrangement
V D J Rearranged DNA
after Oprea & Forrest, 1998
ICANNGA 2001 - An Introduction to the Artificial Immune Systems 13
• Clonal SelectionThe Immune System (X)
Foreign antigens
Proliferation
Differentiation
Plasma cells
Memory cellsSelection
M
M
Antibody
Self-antigen
Self-antigen
Clonal deletion(negative selection)
Clonal deletion(negative selection)
ICANNGA 2001 - An Introduction to the Artificial Immune Systems 14
• Reinforcement Learning and Immune Memory
The Immune System (XI)
Antigen Ag1AntigensAg1, Ag2
Primary Response Secondary Response
Lag
Responseto Ag1
Ant
ibod
y C
once
ntra
tion
Time
Lag
Responseto Ag2
Responseto Ag1
...
...
Cross-ReactiveResponse
...
...
AntigenAg1’
Response toAg1’
Lag
ICANNGA 2001 - An Introduction to the Artificial Immune Systems 15
• Affinity Maturation– somatic hypermutation
– receptor editing
The Immune System (XII)
1ary 2ary 3ary
Affi
nity
(lo
g sc
ale)
Response
ICANNGA 2001 - An Introduction to the Artificial Immune Systems 16
• Self/Nonself Discrimination– repertoire completeness
– co-stimulation
– tolerance
• Positive selection– B- and T-cells are selected as
immunocompetent cells
– Recognition of self-MHC molecules
• Negative selection– Tolerance of self: those cells that recognize
the self are eliminated from the repertoire
The Immune System (XIII)
ICANNGA 2001 - An Introduction to the Artificial Immune Systems 17
• Self/Nonself Discrimination
The Immune System (XIV)
Clonaldeletion
AnergyUnaffected cell
Clonal Expansion Negative SelectionClonal Ignorance
Effector clone
Self-antigen
OR receptor editing
Nonselfantigens
ICANNGA 2001 - An Introduction to the Artificial Immune Systems 18
• Immune Network Theory– The immune system is composed of an enormous and
complex network of paratopes that recognize sets ofidiotopes, and of idiotopes that are recognized by setsof paratopes, thus each element can recognize as wellas be recognized (Jerne, 1974)
• Features (Varela et al., 1988)– Structure
– Dynamics
– Metadynamics
The Immune System (XV)
A n tibody
P a ra tope
Id io top e
ICANNGA 2001 - An Introduction to the Artificial Immune Systems 19
• Immune Network Dynamics
pa
ia
pb
ib
pc
ic
Ag(epitope)
Dynamics of the immune system
Foreign stimulus
Internal image
Anti-idiotypic set
Recognition
ia
px
Non-specific set
The Immune System (XVI)
after Jerne, 1974
ICANNGA 2001 - An Introduction to the Artificial Immune Systems 20
• Pathogen, Antigen, Antibody• Lymphocytes: B- and T-cells• Affinity• 1ary, 2ary and cross-reactive response• Learning and memory
– increase in clone size and affinity maturation
• Self/Nonself Discrimination• Immune Network Theory
The Immune System— Summary —
ICANNGA 2001 - An Introduction to the Artificial Immune Systems 21
• Artificial Immune Systems (AIS)– Remarkable Immune Properties
– Concepts, Scope and Applications
– History of the AIS
– The Shape-Space Formalism
– Representations and Affinities
– Algorithms and Processes
– A Discrete Immune Network Model
– Guidelines to Design an AIS
— Part II —
ICANNGA 2001 - An Introduction to the Artificial Immune Systems 22
• Remarkable Immune Properties– uniqueness
– diversity
– robustness
– autonomy
– multilayered
– self/nonself discrimination
– distributivity
– reinforcement learning and memory
– predator-prey behavior
– noise tolerance (imperfect recognition)
Artificial Immune S ystems (I)
ICANNGA 2001 - An Introduction to the Artificial Immune Systems 23
• Concepts– Artificial immune systems are data manipulation,
classification, reasoning and representationmethodologies, that follow a plausible biologicalparadigm: the human immune system (Starlab)
– An artificial immune system is a computational systembased upon metaphors of the natural immune system(Timmis, 2000)
– The artificial immune systems are composed ofintelligent methodologies, inspired by the naturalimmune system, for the solution of real-world problems(Dasgupta, 1998)
Artificial Immune S ystems (II)
ICANNGA 2001 - An Introduction to the Artificial Immune Systems 24
Artificial Immune Systems (III)• Scope (Dasgupta, 1998):
– Computational methods inspired by immune principles;
– Multi-agent systems based on immunology;
– Self-organized systems based on immunology;
– Immunity-based systems for the development ofcollective behavior;
– Search and optimization methods based onimmunology;
– Immune approaches for artificial life;
– Immune approaches for the security of informationsystems;
– Immune metaphors for machine-learning.
ICANNGA 2001 - An Introduction to the Artificial Immune Systems 25
• Applications– Pattern recognition;
– Function approximation;
– Optimization;
– Data analysis and clustering;
– Machine learning;
– Associative memories;
– Diversity generation and maintenance;
– Evolutionary computation and programming;
– Fault and anomaly detection;
– Control and scheduling;
– Computer and network security;
– Generation of emergent behaviors.
Artificial Immune S ystems (IV)
ICANNGA 2001 - An Introduction to the Artificial Immune Systems 26
Artificial Immune Systems (V)Year Event Activity Organizer/Speaker
1996 IMBS Workshop Y. Ishida
1997 SMC Special Track / Tutorial D. Dasgupta
SMC Special Track D. Dasgupta1998
Book First edited book D. Dasgupta (ed.)
1999 SMC Special Track D. Dasgupta
Workshop D. Dasgupta
GECCO Talk: “Why Does a Computer Need anImmune System”
S. Forrest
Tutorial: “Immunological Computation”SMC
Special TrackD. Dasgupta
2000
SBRNTutorial: “Artificial Immune Systemsand Their Applications”
L. N. de Castro
ICANNGATutorial: “An Introduction to theArtificial Immune Systems”
L. N. de Castro
CECTutorial: “Artificial Immune Systems:An Emerging Technology”
J. I. Timmis2001
Journal(Special Issue)
IEEE-TEC: Special Issue on ArtificialImmune Systems
D. Dasgupta (ed.)
ICANNGA 2001 - An Introduction to the Artificial Immune Systems 27
• How do we mathematically represent immunecells and molecules?
• How do we quantify their interactions orrecognition?
• Shape-Space Formalism (Perelson & Oster, 1979)– Quantitative description of the interactions between cells
and molecules
• Shape-Space (S) Concepts– generalized shape
– recognition through regions of complementarity
– recognition region
– affinity threshold
Artificial Immune S ystems (VI)
ICANNGA 2001 - An Introduction to the Artificial Immune Systems 28
• Recognition Via Regions of Complementarity
Antibody
Antigen
Artificial Immune S ystems (VII)
ICANNGA 2001 - An Introduction to the Artificial Immune Systems 29
• Shape-Space (S)
Artificial Immune S ystems (VIII)
εVε
εVε
εVε
V
× ×
×
×
×
×
×
after Perelson, 1989
ICANNGA 2001 - An Introduction to the Artificial Immune Systems 30
• Representations– Set of coordinates: m = ⟨m1, m2, ..., mL⟩, m ∈ SL ⊆ ℜL
– Ab = ⟨Ab1, Ab2, ..., AbL⟩, Ag = ⟨Ag1, Ag2, ..., AgL⟩• Affinities: related to their distance
– Euclidean
– Manhattan
– Hamming
Artificial Immune S ystems (IX)
∑=
−=L
iii AgAbD
1
2)(
∑=
−=L
iii AgAbD
1
∑=
≠
==L
i
ii AgAbD
1 otherwise0
if1/where/�
ICANNGA 2001 - An Introduction to the Artificial Immune Systems 31
• Affinities in Hamming Shape-Space
Artificial Immune S ystems (X)
Affinity: 6
0 0 1 1 0 0 1 1
1 1 1 0 1 1 0 1
1 1 0 1 1 1 1 0
0 0 1 1 0 0 1 1
1 1 1 0 1 1 0 1
1 1 0 1 1 1 1 0
Affinity: 4
0 0 1 1 0 0 1 1
1 1 1 0 1 1 0 1
1 1 0 1 1 1 1 0
Affinity: 6+24=22
Hamming r-contiguous bit Affinity measure distance rule of Hunt
∑+=i
lH
iDD 2
0 0 1 1 0 0 1 1
1 1 1 0 1 1 0 1
0 0 1 1 0 0 1 1
1 1 1 1 0 1 1 0
0 0 1 1 0 0 1 1
1 1 0 1 1 0 1 1
. . .
Affinity: 6 + 4 + ... + 4 = 32Flipping one string
ICANNGA 2001 - An Introduction to the Artificial Immune Systems 32
• Algorithms and Processes– Main generic algorithms that model specific
immune principles
• Examples– Generation of initial antibody repertoires (Bone
Marrow)
– A Negative selection algorithm
– A Clonal selection algorithm
– Affinity maturation
– Immune network models
Artificial Immune S ystems (XI)
ICANNGA 2001 - An Introduction to the Artificial Immune Systems 33
• Generation of Initial Antibody Repertoires
Artificial Immune S ystems (XII)
An individual genome corresponds to four libraries:
Library 1 Library 2 Library 3 Library 4
A1 A2 A3 A4 A5 A6 A7 A8
A3 D5C8B2
A3 D5C8B2
A3 B2 C8 D5
= four 16 bit segments
= a 64 bit chain
Expressed Ab molecule
B1 B2 B3 B4 B5 B6 B7 B8 C1 C2 C3 C4 C5 C6 C7 C8 D1 D2 D3 D4 D5 D6 D7 D8
after Perelson et al., 1996
ICANNGA 2001 - An Introduction to the Artificial Immune Systems 34
• A Negative Selection Algorithm– store information about the complement of the patterns
to be recognized
Artificial Immune S ystems (XIII)
Selfstrings (S)
Generaterandom strings
(R0)Match Detector
Set (R)
Reject
No
Yes
No
Yes
Detector Set(R)
SelfStrings (S)
Match
Non-selfDetected
Censoring Monitoring phase phase
after Forrest et al., 1994
ICANNGA 2001 - An Introduction to the Artificial Immune Systems 35
• A Clonal Selection Algorithm– the clonal selection principle with applications to
machine-learning, pattern recognition and optimization
Artificial Immune S ystems (XIV)
Ab (1)
(2)
(3)
(4)
Ab{ d}(8)
Maturate
Select
Clone
Ab{n}
C
C*
Re-select
f
f
(7)
(6)
(5)
Ab{ n}
after de Castro & Von Zuben, 2001a
ICANNGA 2001 - An Introduction to the Artificial Immune Systems 36
• Somatic Hypermutation– Hamming shape-space with an alphabet of length 8
– Real-valued vectors: inductive mutation
Artificial Immune S ystems (XV)
3 1 2 4 6 1 8 7
3 1 6 4 2 1 8 7
Single-point mutation
Original string
Mutated string
Bits to be swapped
ICANNGA 2001 - An Introduction to the Artificial Immune Systems 37
• Affinity Proportionate Hypermutation
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
D*
α
ρ = 5
ρ = 10
ρ = 20
Artificial Immune S ystems (XVI)
0 20 40 60 80 100 120 140 160 180 200
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
α
Iterations
after de Castro & Von Zuben, 2001a after Kepler & Perelson, 1993
ICANNGA 2001 - An Introduction to the Artificial Immune Systems 38
• A Discrete Immune Network Model: aiNet
Artificial Immune S ystems (XVII)
1. For each antigenic pattern Agi,1.1 Clonal selection: Apply the pattern recognition version
of CLONALG that will return a matrix of memoryclones for Agi;
1.2 Apoptosis: Eliminate all memory clones whose affinitywith antigen are below a threshold;
1.3 Inter-cell affinity: Determine the affinity between allclones generated for Agi;
1.4 Clonal Suppression: Eliminate those clones whoseaffinities are inferior to a pre-specified threshold;
1.5 Total repertoire: Concatenate the clone generated forantigen Agi with all network cells
2. Inter-cell affinity: Determine the affinity between all networkcells;
3. Network suppression: Eliminate all aiNet cells whose affinitiesare inferior to a pre-specified threshold.
ICANNGA 2001 - An Introduction to the Artificial Immune Systems 39
• Guidelines to Design an AIS
Artificial Immune S ystems (XVIII)
1. Problem definition2. Mapping the real problem into the AIS domain
2.1 Defining the types of immune cells and moleculesto be used
2.2 Deciding the immune principle(s) to be used in thesolution
2.3 Defining the mathematical representation for theelements of the AIS
2.4 Evaluating the interactions among the elements ofthe AIS (dynamics)
2.5 Controlling the metadynamics of the AIS3. Reverse mapping from AIS to the real problem
ICANNGA 2001 - An Introduction to the Artificial Immune Systems 40
• Examples of Artificial Immune Systems– Network Intrusion Detection by Hofmeyr &
Forrest (2000)
– aiNet: An Artificial Immune Network Model byde Castro & Von Zuben (2001)
• A Tour on the Clonal Selection Algorithm(CLONALG) and aiNet
• Discussion and Future Trends
— Part III —
ICANNGA 2001 - An Introduction to the Artificial Immune Systems 41
• Computer Security– direct metaphor
– virus and network intrusion detection
• Network Intrusion Detection by Hofmeyr &Forrest (2000)– Rationale: protect a computer network against illegal
users
– Basic cell type: detector that can assume several states,such as thymocyte, naive B-cell, memory B-cell
– Representation: Hamming shape-space and r-contiguous bits rule
Examples of AIS (I)
ICANNGA 2001 - An Introduction to the Artificial Immune Systems 42
Examples of AIS (II)Immune System Artificial Immune System
Receptor Bitstring
Lymphocyte (B- and T-cell) Detector
Memory cell Memory detector
Pathogen Non-self bitstring
Binding Approximate string matching
Circulation Mobile detectors
MHC Representation parameters
Cytokines Sensitivity level
Tolerization Distributed negative selection
Co-stimulation: signal one A match exceeding theactivation threshold
Co-stimulation: signal two Human operator
Lymphocyte cloning Detector replication
ICANNGA 2001 - An Introduction to the Artificial Immune Systems 43
• Life-cycle of a detector
Examples of AIS (III)
Randomly created
Immature
Mature & Naive
Death
Activated
Memory
No match duringtolerization
010011100010.....001101
Exceed activationthreshold
Don’t exceedactivation threshold
No costimulation Costimulation
Match
Match duringtolerization
after Hofmeyr & Forrest, 2000
ICANNGA 2001 - An Introduction to the Artificial Immune Systems 44
• aiNet: An Artificial Immune Network Model– The aiNet is a disconnected graph composed of a set of
nodes, called cells or antibodies, and sets of node pairscalled edges with a number assigned called weight, orconnection strength, specified to each connected edge(de Castro & Von Zuben, 2001)
Examples of AIS (IV)
1
2 0.11
0.4
0.1
0.3 0.25
4
5
68 7 0.1 0.1
3 0.06
0.05
ICANNGA 2001 - An Introduction to the Artificial Immune Systems 45
Examples of AIS (V)• Rationale:
– To use the clonal selection principle together with theimmune network theory to develop an artificial networkmodel using a different paradigm from the ANN.
• Applications:– Data compression and analysis.
• Properties:– Knowledge distributed among the cells
– Competitive learning (unsupervised)
– Constructive model with pruning phases
– Generation and maintenance of diversity
ICANNGA 2001 - An Introduction to the Artificial Immune Systems 47
Discussion
• Growing interest for the AIS
• Biologically Motivated Computing– utility and extension of biology
– improved comprehension of natural phenomena
• Example-based learning, where differentpattern categories are represented byadaptive memories of the system
• Strongly related to other intelligentapproaches, like ANN, EC, FS, DNAComputing, etc.
ICANNGA 2001 - An Introduction to the Artificial Immune Systems 48
• The proposal of a general frameworkin which to design AIS
• Relate AIS with ANN, EC, FS, etc.– Similarities and differences
– Equivalencies
• Applications– Optimization
– Data Analysis
– Machine-Learning
– Pattern Recognition
• Hybrid algorithms
Future Trends
ICANNGA 2001 - An Introduction to the Artificial Immune Systems 49
• Dasgupta, D. (Ed.) (1998), Artificial Immune Systems and TheirApplications, Springer-Verlag.
• De Castro, L. N., & Von Zuben, F. J., (2001a), “Learning andOptimization Using the Clonal Selection Principle”, submitted to theIEEE Transaction on Evolutionary Computation (Special Issue on AIS).
• De Castro, L. N. & Von Zuben, F. J. (2001), "aiNet: An Artificial ImmuneNetwork for Data Analysis", Book Chapter in Data Mining: A HeuristicApproach, Hussein A. Abbass, Ruhul A. Sarker, and Charles S. Newton(Eds.), Idea Group Publishing, USA.
• Forrest, S., A. Perelson, Allen, L. & Cherukuri, R. (1994), “Self-NonselfDiscrimination in a Computer”, Proc. of the IEEE Symposium onResearch in Security and Privacy, pp. 202-212.
• Hofmeyr S. A. & Forrest, S. (2000), “Architecture for an ArtificialImmune System”, Evolutionary Computation, 7(1), pp. 45-68.
• Jerne, N. K. (1974a), “Towards a Network Theory of the ImmuneSystem”, Ann. Immunol. (Inst. Pasteur) 125C, pp. 373-389.
• Kepler, T. B. & Perelson, A. S. (1993a), “Somatic Hypermutation in BCells: An Optimal Control Treatment”, J. theor. Biol., 164, pp. 37-64.
• Klein, J. (1990), Immunology, Blackwell Scientific Publications.
References (I)
ICANNGA 2001 - An Introduction to the Artificial Immune Systems 50
References (II)• Nossal, G. J. V. (1993a), “Life, Death and the Immune System”, Scientific
American, 269(3), pp. 21-30.• Oprea, M. & Forrest, S. (1998), “Simulated Evolution of Antibody Gene
Libraries Under Pathogen Selection”, Proc. of the IEEE SMC’98.• Perelson, A. S. (1989), “Immune Network Theory”, Imm. Rev., 110, pp. 5-36.• Perelson, A. S. & Oster, G. F. (1979), “Theoretical Studies of Clonal Selection:
Minimal Antibody Repertoire Size and Reliability of Self-NonselfDiscrimination”, J. theor.Biol., 81, pp. 645-670.
• Perelson, A. S., Hightower, R. & Forrest, S. (1996), “Evolution and SomaticLearning in V-Region Genes”, Research in Immunology, 147, pp. 202-208.
• Starlab, URL: http://www.starlab.org/genes/ais/• Timmis, J. (2000), Artificial Immune Systems: A Novel Data Analysis
Technique Inspired by the Immune Network Theory, Ph.D. Dissertation,Department of Computer Science, University of Whales, September.
• Tizard, I. R. (1995), Immunology An Introduction, Saunders College Publishing,4th Ed.
• Varela, F. J., Coutinho, A. Dupire, E. & Vaz, N. N. (1988), “CognitiveNetworks: Immune, Neural and Otherwise”, Theoretical Immunology, Part II,A. S. Perelson (Ed.), pp. 359-375.