» next  » last 

Jump to:

Artificial Immune Systems and Negative Selection

1. Introduction

The aims of this article are twofold. Firstly, the article gives an introduction to Artificial Immune Systems (AIS), their operation and biological inspiration. Secondly it is hoped that the component framework developed here will provide a sound foundation for Delphi developers wishing to exploit AIS technology.

We start by giving a brief overview of the human immune system, its distributed architecture and the in built mechanisms for detecting intruders by self/non self discrimination. The next section lists the computational aspects of the immune system which form the basis of data structures and algorithms wemight use to represent it. The architecture of an artificial immune system is outlined in the next section, followed by details of the components that were developed for this article. These components are used in two demo applications which exploit the capabilities of AIS in very different ways. Finally we conclude with a discussion of how the AIS approach differs from other AI systems.

2. The Human Immune System

The purpose of the human immune system is twofold: It must be able to detect and categorise foreign bodies (antigens) and must also be able to destroy or neutralise them to protect us from their effects. We will concentrate on the first of these, the ability to detect and categorise antigens. The term antigen refers to anything which is not part of the body itself, including bacteria, viruses and suchlike. The job of the immune system is to differentiate between antigens and the body itself, initiating the appropriate immune response for foreign bodies, whilst leaving everything else untouched. This is known as "self/ non-self discrimination".

The main component of the immune system is the lymphocyte. Countless millions of these circulate through the bloodstream and within the tissues of the body. Lymphocytes are able to recognise non-self (i.e. antigens) by binding to them with chemical bonds as illustrated in figure 1. The more complimentary the antigen and lymphocyte the higher the likelihood of a bond forming. The probability of a bond forming is known as affinity - high affinity is a high likelihood of a bond forming.

An antigen binding to a lymphocyte with an exact match
Figure 1: An antigen binding to a lymphocyte with an exact match

The binding shown in figure 1 is an exact match. However, to ensure that the immune system can recognise as many antigens as possible an exact match is not required. A match can occur if a given number of contiguous features complement each other. The number of features required to bind before a match can be made is known as the affinity threshold. Figure 2 shows a lymphocyte and antigen which match, though not perfectly - there are four contiguous matching bonds (numbers 2 to 5). Figure 3 shows a lymphocyte which does not match the antigen as there are no sets of contiguous features long enough to cause a match (assuming our affinity threshold is greater than two).

An antigen binding to a lymphocyte with a partial match
Figure 2: An antigen binding to a lymphocyte with a partial match

An antigen and lymphocyte which do not match
Figure 3: An antigen and lymphocyte which do not match

There are two types of lymphocytes: B cells and T cells. We are only interested in T cells here. All lymphocytes have three stages in their lifecycle.

  • Naive lymphocytes have not yet been "trained" to recognise a specific antigen
  • Effector lymphocytes are active within the immune system
  • Memory lymphocytes exist only to store patterns associated with antigens that have invaded the body in the past

Lymphocytes are "born" in the lymphoid organs through the division of existing lymphocyte cells. An unusually large level of mutation takes place as these lymphocytes divide (known as Somatic Hypermutation). This means that "child" lymphocytes are different, though similar to their "parents". This mutation allows the immune system to learn to categorise new threats. The process is known as clonal expansion.

During clonal expansion it is possible that new lymphocytes will be created which incorrectly categorise proteins in the body (self) as antigens (non-self). These must not be allowed to circulate around the body as they will cause the immune system to attack the body's tissues. Therefore, lymphocytes are subjected to a negative selection process in which they are tested to ensure they do not categorise self proteins. Lymphocytes which fail the test are destroyed, those which pass are allowed to circulate around the body.

When a new antigen invades the body only a few lymphocytes are able to recognise it. The body must quickly produce more lymphocytes with high affinity to the antigen, before it has had time to damage the body too much. For this reason lymphocytes with a high affinity for antigens are encouraged to reproduce more, that is to say that recognition stimulates cells to reproduce and mutate. Cells which mutate to find a better match with the target antigen are encouraged to divide more. This process creates a large number of lymphocytes which are specific to the antigen.

The process of going from a reasonably low affinity to an antigen in a few lymphocytes to produce a large population of high affinity lymphocytes is called Affinity Maturation. This process uses reproduction and mutation to create new cells and favours higher affinity (fitter) cells for reproduction. Essentially, affinity maturation is a Darwinian evolutionary process.

Once an antigen has been destroyed, the immune system will remember it, in case it enters the body in future. This memory comes about through memory lymphocytes, which are created in the same was as effector lymphocytes but have a longer life span. Memory cells serve as a reminder of antigen patterns which have posed a threat in the past. Because memory cells have a high affinity for a given antigen they are more likely to reproduce. Thus a subset of the lymphocyte population will continue to have a high affinity for the remembered antigen.

So, the immune system is diverse in that a large number of different antigens can be detected, protecting the body from just about anything which might invade it. The negative selection process ensures that lymphocytes do not attack the body itself by checking all new lymphocytes against a set of "self" proteins. It can quickly respond to a new antigen because lymphocytes with high affinities are encouraged to reproduce more. The immune system is also able to remember antigens it has dealt with in the past, by keeping a population of memory lymphocytes in circulation.

3. Properties of the Immune System

Some properties of the immune system, computational and otherwise, are enumerated below.
  • Recognition: Self/ non-self categorisation takes place based on intercellular binding
  • Diversity: A diverse set of lymphocytes is created, partly by genetic process, to ensure that cells exist which can bind to any antigen - known or unknown
  • Learning: Clonal expansion, during the first encounter with an antigen allows the immune system to learn, by example, to recognise specific antigens
  • Memory: Memory cells store the patterns associated with antigens encountered in the past
  • Distributed Detection: The immune system is inherently distributed. Lymphocytes and continually recirculated around the body where they encounter antigenic challenges and stimulate the appropriate immune response
  • Self Regulation: Regulation takes place either locally or on a system wide basis, depending on the situation
  • Threshold Mechanism: Immune system response takes place above a threshold, dictated by the strength of the chemical binding between lymphocyte and antigen
  • Dynamic Protection: Affinity maturation (clonal expansion and somatic hyper-mutation) allows the generation of high affinity lymphocytes. This dynamically balances exploration versus exploitation in immunity. Dynamic protection increases the coverage of the immune system over time
  • Probabilistic Detection:The cross reaction in immune response is a stochastic process. Detection is approximate, so one lymphocyte may bind to several antigens

» next  » last 

Jump to: