Engelbrecht Andries P. Computational Intelligence: An Introduction

Подождите немного. Документ загружается.

418 18. Natural Immune System

Celldevelops into develops into LymphocytesPhagocytes

Bone Marrow

Creates

Macrophages

NeutrophilsMonocytes

BŦCells

TŦCells

Figure 18.2 White Cell Types

internal structure of a cell. MHC-molecules are grouped into two classes: Type I and

Type II. MHC-molecules of Type I is on the surface of any cell and MHC-molecules

of Type II mainly on the surface of B-Cells [678]. There are two types of T-Cells:

The Helper-T-Cell and Natural-Killer-T-Cell. Each of these types of lymphocytes are

describedindetailbelow.

Immature Mature

Memory

Annihilated

Figure 18.3 Life Cycle of A Lymphocyte

The B-Cell

B-Cells are created in the bone marrow with monomeric IgM-receptors on their sur-

faces. A monomeric receptor is a chemical compound that can undergo a chemical

18.3 The White Cells 419

Partitioning antigen

Antigen Receptor

BŦCell

MHCŦII

Peptides

Mature TŦCell

Lymphokines

Plasma cell

bod

Figure 18.4 B-Cell Develops into Plasma Cell, Producing Antibodies

reaction with other molecules to form larger molecules. In contrast to T-Cells, B-Cells

leave the bone marrow as mature lymphocytes. B-Cells mostly exist in the spleen and

tonsils. It is in the spleen and tonsils that the B-Cells develop into plasma cells after

the B-Cells come into contact with antigens. After developing into plasma cells, the

plasma cells produce antibodies that are eﬀective against antigens [582]. The B-Cell

has antigen-speciﬁc receptors and recognizes in its natural state the antigens. When

contact is made between B-Cell and antigen, clonal proliferation on the B-Cell takes

place and is strengthened by Helper-T-Cells (as explained the next subsection). Dur-

ing clonal proliferation two types of cells are formed: plasma cells and memory cells.

The function of memory cells is to proliferate to plasma cells for a faster reaction to

frequently encountered antigens and produce antibodies for the antigens. A plasma

cell is a B-Cell that produces antibodies.

The Helper-T-Cell (HTC)

When a B-Cell’s receptor matches an antigen, the antigen is partitioned into peptides

(as shown in Figure 18.4). The peptides are then brought to the surface of the B-Cell

by an MHC-molecule of Type II. Macrophages also break down antigen and the broken

down antigen is brought to the surface of the macrophage by an MHC-molecule of Type

II. The HTC binds to the MHC-molecule on the surface of the B-Cell or macrophage

420 18. Natural Immune System

and proliferates or suppresses the B-Cell response to the partitioned cell, by secreting

lymphokines. This response is known as the primary response. When the HTC bounds

to the MHC with a high aﬃnity, the B-Cell is proliferated. The B-Cell then produces

antibodies with the same structure or pattern as represented by the peptides. The

production of antibodies is done after a cloning process of the B-Cell.

When the HTC does not bind with a high aﬃnity, the B-Cell response is suppressed.

Aﬃnity is a force that causes the HTC to elect a MHC on the surface of the B-Cell

with which the HTC has a stronger binding to unite, rather than with another MHC

with a weaker binding. A higher aﬃnity implies a stronger binding between the HTC

and MHC. The antibodies then bind to the antigens’ epitopes that have the same

complementary structure or pattern. Epitopes are the portions on an antigen that

are recognized by antibodies. When a B-Cell is proliferated enough, i.e. the B-Cell

frequently detects antigens, it goes into a memory status, and when it is suppressed

frequently it becomes annihilated and replaced by a newly created B-Cell. The immune

system uses the B-Cells with memory status in a secondary response to frequently

seen antigens of the same structure. The secondary response is much faster than the

primary response, since no HTC signal or binding to the memory B-Cell is necessary

for producing antibodies [678].

Peptides

MHCŦI

Cell dies

Natu

lŦKill

rŦTŦ

Antigen

Macrophage

Partitioning antigen

Figure 18.5 Macrophage and NKTC

The Natural-Killer-T-Cell (NKTC)

The NKTC binds to MHC-molecules of Type I (as illustrated in Figure 18.5). These

MHC-molecules are found on all cells. Their function is to bring to light any viral

18.4 Immunity Types 421

proteins from a virally infected cell. The NKTC then binds to the MHC-molecule of

Type I and destroys not only the virally infected cell but also the NKTC itself [678].

18.4 Immunity Types

Immunity can be obtained either naturally or artiﬁcially. In both cases immunity can

be active or passive. This section discusses the diﬀerent types of immunity.

Active naturally-obtained immunity: Due to memory-cells, active naturally-

obtained immunity is more or less permanent. It develops when the body gets infected

or receives foreign red blood cells and actively produces antibodies to deactivate the

antigen [582].

Passive naturally-obtained immunity: Passive naturally-obtained immunity is

short-lived since antibodies are continuously broken down without creation of new

antibodies. New antibodies are not created because the antigens did not activate

the self immune system. The immunity type develops from IgG-antibodies that are

transplanted from the mother to the baby. The secreted IgA-antibodies in mothers-

milk are another example of this immunity type and protect the baby from any antigens

with which the mother came into contact [582].

Active artificially-obtained immunity: Active artiﬁcially-obtained immunity

develops when dead organisms or weakened organisms are therapeutically applied.

The concept is that special treated organisms keep their antigens without provoking

illness-reactions [582].

Passive artificially-obtained immunity: Passive artiﬁcially-obtained immunity

is obtained when a speciﬁc antibody that was produced by another human or animal,

is injected into the body for an emergency treatment. Immunity is short-lived, since

the immune system is not activated [582].

18.5 Learning the Antigen Structure

Learning in the immune system is based on increasing the population size of those lym-

phocytes that frequently recognize antigens. Learning by the immune system is done

by a process known as aﬃnity maturation. Aﬃnity maturation can be broken down

into two smaller processes namely, a cloning process and a somatic hyper-mutation

process. The cloning process is more generally known as clonal selection,whichisthe

proliferation of the lymphocytes that recognize the antigens.

422 18. Natural Immune System

The interaction of the lymphocyte with an antigen leads to an activation of the lym-

phocyte where upon the cell is proliferated and grown into a clone. When an antigen

stimulates a lymphocyte, the lymphocyte not only secretes antibodies to bind to the

antigen but also generates mutated clones of itself in an attempt to have a higher

binding aﬃnity with the detected antigen. The latter process is known as somatic

hyper-mutation. Thus, through repetitive exposure to the antigen, the immune sys-

tem learns and adapts to the shape of the frequently encountered antigen and moves

from a random receptor creation to a repertoire that represents the antigens more pre-

cisely. Lymphocytes in a clone produce antibodies if it is a B-Cell and secrete growth

factors (lymphokines) in the case of an HTC.

Since antigens determine or select the lymphocytes that need to be cloned, the process

is called clonal selection [582]. The ﬁttest clones are those which produce antibodies

that bind to antigen best (with highest aﬃnity). Since the total number of lymphocytes

in the immune system is regulated, the increase in size of some clones decreases the

size of other clones. This leads to the immune system forgetting previously learned

antigens. When a familiar antigen is detected, the immune system responds with

larger cloning sizes. This response is referred to as the secondary immune response

[678]. Learning is also based on decreasing the population size of those lymphocytes

that seldom or never detect any antigens. These lymphocytes are removed from the

immune system. For the aﬃnity maturation process to be successful, the receptor

molecule repository needs to be as complete and diverse as possible to recognize any

foreign shape [678].

18.6 The Network Theory

The network theory was ﬁrst introduced by Jerne [416, 677], and states that B-Cells

are interconnected to form a network of cells. When a B-Cell in the network responds

to a foreign cell, the activated B-Cell stimulates all the other B-Cells to which it is

connected in the network. Thus, a lymphocyte is not only stimulated by an antigen,

but can also be stimulated or suppressed by neighboring lymphocytes, i.e. when a

lymphocyte reacts to the stimulation of an antigen, the secretion of antibodies and

generation of mutated clones (see Section 18.5) stimulate the lymphocyte’s immediate

neighbors. This implies that a neighbor lymphocyte can then in turn also react to

the stimulation of the antigen-stimulated lymphocyte by generating mutated clones,

stimulating the next group of neighbors, etc.

18.7 The Danger Theory

The danger theory was introduced by Matzinger [567, 568] and is based on the co-

stimulated model of Laﬀerty and Cunningham [497]. The main idea of the danger

theory is that the immune system distinguishes between what is dangerous and non-

dangerous in the body. The danger theory diﬀers from the classical view in that the

immune system does not respond to all foreign cells, but only to those foreign cells

18.7 The Danger Theory 423

Partitioning antigen

Antigen Receptor

BŦCell

MHCŦII

Peptides

Lymphokines

Plasma cell

Antibodies

APC

Stress signal

antigen

Mature TŦCell

CoŦstimulation

Necrotic cell death

Figure 18.6 Co-Stimulation of T-Cell by an APC

that are harmful or dangerous to the body. A foreign cell is seen to be dangerous

to the body if it causes body cells to stress or die. Matzinger gives two motivational

reasons for deﬁning the new theory, which is that the immune system needs to adapt

to a changing self and that the immune system does not always react on foreign or

non-self.

Although cell death is common within the body, the immune system only reacts to

those cell deaths that are not normal programmed cell death (apoptosis), i.e. non-

apoptotic or necrotic deaths. When a cell is infected by a virus, the cell itself will

send out a stress signal (known as signal 0) of necrotic death to activate the antigen

presenting cells (APCs) (as illustrated in Figure 18.6). Thus, co-stimulation of an

APC to a helper T-Cell is only possible if the APC was activated with a danger

or stress signal. Therefore, the neighboring cells of an APC determines the APC’s

424 18. Natural Immune System

state. Hereon the immune reaction process is as discussed within the classical view

(see Section 18.1), where now mature helper T-Cells are presented with a peptide

representation of the antigen and co-stimulated by an activated APC.

The diﬀerent types of signals from a dying or stressed cell is unknown. According

to Matzinger these signals could either be deﬁned as the sensing of a certain protein

within a cell that leaked after the cell’s death or an unexpected connection lost between

connected cells after one of the cells died. Thus, if none of the above signals are ﬁred

by a cell, no immune response will be triggered by an antigen to activate the antigen

presenting cells (APCs).

Thus, from a danger immune system perspective a T-Cell only needs to be able to

diﬀerentiate APCs from any other cells. If an APC activated a T-Cell through co-

stimulation, then only will the immune system respond with a clonal proliferation of

the B-Cell (as discussed in Section 18.3.1). The B-Cell will then secrete antibodies to

bind with the dangerous antigen instead of binding to all foreign harmless antigen.

18.8 Assignments

1. Identify the main diﬀerence between the classicial view of the NIS

(a) and network theory,

(b) and danger theory.

2. Discuss the merit of the following statement: “The lymphocytes in the classical

view perform a pattern matching function. ”

3. At this point, discuss how the principles of a NIS can be used to solve real-world

problems where anomalies need to be detected, such as fraud.

4. Discuss the merit of the following statement: “A model of the NIS (based on the

classical view) can be used as a classifier.”

Chapter 19

Artiﬁcial Immune Models

Chapter 18 discussed the diﬀerent theories with regards to the functioning and orga-

nizational behavior of the natural immune system (NIS). These theories inspired the

modeling of the NIS into an artiﬁcial immune system (AIS) for application in non-

biological environments. Capabilities of the NIS within each theory, are summarised

below:

• The NIS only needs to know the structure of self/normal cells.

• The NIS can distinguish between self and foreign/non-self cells.

• A foreign cell can be sensed as dangerous or non-dangerous.

• Lymphocytes are cloned and mutated to learn and adapt to the structure of the

encountered foreign cells.

• The build-up of a memory on the learned structures of the foreign cells.

• A faster secondary response to frequently encountered foreign cells, due to the

built-up memory.

• The cooperation and co-stimulation among lymphocytes to learn and react to

encountered foreign cells.

• The formation of lymphocyte networks as a result of the cooperation and co-

stimulation among lymphocytes.

This chapter discusses some of the existing AIS models. More detail is provided on

the most familiar AIS models. These models are either based on or inspired by the

capabilities of the NIS (as summarised above) and implement some or all of the basic

AIS concepts as listed in Section 19.1. The chapter contains a section for each of

the diﬀerent theories in immunology as discussed in Chapter 18. In each of these

sections, the AIS models that are based on or inspired by the applicable theory, are

discussed. The rest of the chapter is organised as follows: Section 19.1 lists the basic

concepts of an artiﬁcial immune system. A basic AIS algorithm is proposed and given

in pseudo code. The diﬀerent parts of the basic AIS pseudo code is brieﬂy discussed.

Section 19.2 discusses the AIS models based on or inspired by the classical view of

the natural immune system. Section 19.3 gives an algorithm in pseudo code that is

inspired by the clonal selection theory of the natural immune system. Other clonal

selection inspired models are also discussed. Section 19.4 discusses the AIS models

based on or inspired by the network theory of the natural immune system. Section 19.5

discusses the AIS models based on or inspired by danger theory. Section 19.6 concludes

Computational Intelligence: An Introduction, Second Edition A.P. Engelbrecht

2007 John Wiley & Sons, Ltd

425

426 19. Artificial Immune Models

the chapter by giving a brief overview on some of the problem domains to which the

artiﬁcial immune system has been successfully applied.

19.1 Artiﬁcial Immune System Algorithm

The capabilities of the NIS summarized above imply that it is mainly the inner working

and cooperation between the mature T-Cells and B-Cells that is responsible for the

secretion of antibodies as an immune response to antigens. The T-Cell becomes mature

in the thymus. A mature T-Cell is self-tolerant, i.e. the T-Cell does not bind to self

cells. The mature T-Cell’s ability to discriminate between self cells and non-self cells

makes the NIS capable of detecting non-self cells. When a receptor of the B-Cell binds

to an antigen, the antigen is partitioned and then brought to the surface with an

MHC-molecule. The receptor of the T-Cell binds with a certain aﬃnity to the MHC-

molecule on the surface of the B-Cell. The aﬃnity can be seen as a measurement to

the number of lymphokines that must be secreted by the T-Cell to clonally proliferate

the B-Cell into a plasma cell that can produce antibodies. The memory of the NIS

on frequently detected antigen is built-up by the B-Cells that frequently proliferate

into plasma cells. Thus, to model an AIS, there are a few basic concepts that must be

considered:

• There are trained detectors (artiﬁcial lymphocytes) that detect non-self patterns

with a certain aﬃnity.

• The artiﬁcial immune system may need a good repository of self patterns or

self and non-self patterns to train the artiﬁcial lymphocytes (ALCs) to be self-

tolerant.

• The aﬃnity between an ALC and a pattern needs to be measured. The measured

aﬃnity indicates to what degree an ALC detects a pattern.

• To be able to measure aﬃnity, the representation of the patterns and the ALCs

need to have the same structure.

• The aﬃnity between two ALCs needs to be measured. The measured aﬃnity

indicates to what degree an ALC links with another ALC to form a network.

• The artiﬁcial immune system has memory that is built-up by the artiﬁcial lym-

phocytes that frequently detect non-self patterns.

• When an ALC detects non-self patterns, it can be cloned and the clones can be

mutated to have more diversity in the search space.

Using the above concepts as a guideline, the pseudo code in Algorithm 19.1 is a

proposal of a basic AIS. Each of the algorithm’s parts are brieﬂy explained next.

1. Initializing C and determining D

: The population C can either be populated

with randomly generated ALCs or with ALCs that are initialized with a cross

section of the data set to be learned. If a cross section of the data set is used to

initialize the ALCs, the complement of the data set will determine the training

set D

. These and other initialization methods are discussed for each of the AIS

models in the sections to follow.

19.1 Artificial Immune System Algorithm 427

Algorithm 19.1 Basic AIS Algorithm

Initialize a set of ALCs as population C;

Determine the antigen patterns as training set D

;

while some stopping condition(s) not true do

for each antigen pattern z

∈ D

Select a subset of ALCs for exposure to z

, as population S⊆C;

for each ALC, x

∈Sdo

Calculate the antigen affinity between z

and x

;

end

Select a subset of ALCs with the highest calculated antigen affinity as

population H⊆S;

Adapt the ALCs in H with some selection method, based on the calculated

antigen affinity and/or the network affinity among ALCs in H;

Update the stimulation level of each ALC in H;

end

2. Stopping condition for the while-loop: In most of the discussed AIS models,

the stopping condition is based on convergence of the ALC population or a preset

number of iterations.

3. Selecting a subset, S,ofALCs: The selected subset S can be the entire set

P or a number of randomly selected ALCs from P. Selection of S can also be

based on the stimulation level (as discussed below).

4. Calculating the antigen affinity: The antigen aﬃnity is the measurement

of similarity or dissimilarity between an ALC and an antigen pattern. The most

commonly used measures of aﬃnity in existing AIS models are the Euclidean

distance, r-continuous matching rule, hamming distance and cosine similarity.

5. Selecting a subset, H,ofALCs: In some of the AIS models, the selection of

highest affinity ALCs is based on a preset aﬃnity threshold. Thus, the selected

subset H can be the entire set S, depending on the preset aﬃnity threshold.

6. Calculating the network affinity: This is the measurement of affinity be-

tween two ALCs. The diﬀerent measures of network aﬃnity are the same as those

for antigen affinity. A preset network aﬃnity threshold determines whether two

or more ALCs are linked to form a network.

7. Adapting the ALCs in subset H: Adaptation of ALCs can be seen as the

maturation process of the ALC, supervised or unsupervised. Some of the se-

lection methods that can be used are negative selection (or positive selection),

clonal selection and/or some evolutionary technique with mutation operators.

ALCs that form a network can inﬂuence each other to adapt to an antigen.

These selection methods are discussed for each of the discussed AIS models in

the chapter.

8. Updating the stimulation of an ALC: The stimulation level is calculated in

diﬀerent ways in existing AIS models. In some AIS models, the stimulation level

is seen as the summation of antigen aﬃnities, which determines the resource