Artificial Immune Systems

Size: px
Start display at page:

Download "Artificial Immune Systems"

Transcription

1 Artificial Immune Systems Dr. Mario Pavone Department of Mathematics & Computer Science University of Catania

2 Biological Immune System (1/4) Immunology is the study of the defense mechanisms that confer resistance against diseases Immune System (IS) is the main responsible to protect the organism against the attack from external microorganisms, that might cause diseases (pathogen: viruses, funguses, bacteria and parasites) The biological IS has to assure recognition of each potentially dangerous molecule or substance (antigen Ag) Antigen is any molecule that can stimulate the IS IS, first,distinguishes between the cells of the organism (self ) and those that do not belong to it (nonself ), and then it eliminates the dangerous or extraneous cells and those that have been infected, soastoavoidorblockthedisease. Mario Pavone, IBM-KAIST Bio-Computing Research Center p.7/41

3 Biological Immune System (2/4) The process of distinguishing between what cells belong and does not belong to the organism is called self/nonself discrimination Bone marrow: organswherethebloodcellsaregenerated and developed Thymus: organs where a class of immune cells migrates and matures Lymphocytes: are white blood cells,specialized in the recognition of pathogens Bcells,whichdevelopwithinthebone marrow Tcells,whichmigrateanddevelopwithinthethymus There are two kinds of T cells: cytotoxic (or killer) T cells and helper T cells Mario Pavone, IBM-KAIST Bio-Computing Research Center p.8/41

4 Biological Immune System (3/4) Both lymphocytes present receptor molecules on their surfaces, which are responsible to recognize the Ag TCR and BCR (or antibody) The purposes of the Ab are: recognize and bind with ad Ag perform an effector function While the antibody can recognize and bind only antigens free in solution, TCRs can only recognize and bind with antigens presented by self molecules major histocompatibility complex - MHC Mario Pavone, IBM-KAIST Bio-Computing Research Center p.9/41

5 Biological Immune System (4/4) Each receptor has a specific shape and can only react with a certain antigen Recognition in the immune system is based on shape complementary If a receptor binds an antigen then the cell will be activated through the binds, a virus may be inactivated The binding between receptors and antigens, trigger the immune response Mario Pavone, IBM-KAIST Bio-Computing Research Center p.10/41

6 Bcellreceptor Antibody(1/2) Mario Pavone, IBM-KAIST Bio-Computing Research Center p.11/41

7 Bcellreceptor Antibody(2/2) Mario Pavone, IBM-KAIST Bio-Computing Research Center p.12/41

8 Tcellreceptor Mario Pavone, IBM-KAIST Bio-Computing Research Center p.13/41

9 Humoral Immune Response An immune response is provoked when the immune system encounters a foreign molecule A number of receptors will be produced by the immune system in response to the infection, which will help to eliminate the antigen All receptors that better recognize the antigen, willbeselectedtohavelong life spans (Immune Memory) The production of cells with longer expected lifetime assures the organism a higher specific responsiveness to that antigenic pattern Primary Response: theantigenisrecognized and the memory is developed Secondary Response: arapidandmoreabundantproductionofantibodiesis obtained from the stimulation of cells already specialized and present as memory cells, whenthesameantigenisencounteredagain This means that the body is ready to combat any re-infection Mario Pavone, IBM-KAIST Bio-Computing Research Center p.14/41

10 Primary and Secondary Immune Responses Mario Pavone, IBM-KAIST Bio-Computing Research Center p.15/41

11 What is an Artificial Immune System? Artificial Immune Systems (AIS) represent a field of biologically inspired computing that attempts to exploit theories, principles, and concepts of modern immunology to design immune system based applications to solve problems in science and engineering AIS are population based, as a typical evolutionary algorithm (EA) Computational Intelligence and EAs are optimization methods based on an evolutionary metaphor that showed effective in solving difficult problems Mario Pavone, IBM-KAIST Bio-Computing Research Center p.4/41

12 Artificial Immune Systems Computational Immunology: AISisamodeloftheimmune system that can be used by immunologists for explanation, experimentation and prediction activities that would be difficult or impossible in laboratory experiments AIS is an abstraction of one or more immunological processes to tackle complex problem domains Mario Pavone, IBM-KAIST Bio-Computing Research Center p.6/41

13 Processes of the Immune System Clonal Selection Proliferation and differentiation of the cells, which better recognize nonself entities Negative Selection Eliminating of T cells in the thymus, which recognize self entities Immune Network How do the cells of the immune system interact with each other? Danger Theory Why there is an immune response for harmful self entities, and not for harmless nonself? Mario Pavone, IBM-KAIST Bio-Computing Research Center p.16/41

14 Clonal Selection The focus of the clonal selection is on how B cells can adapt to match and kill invaders When a Bcellmatchesanantigen,thiscauses a B cell to be cloned, proportionallytoitsmatch squality Among all possible cells present in the organism, those who are able to recognize the antigen, will start to proliferate by duplication (cloning) [Burnet, Cambridge University Press, 1959] Clonal Expansion phase: istheprocesswherethebcells produce many clones, when they are activated by binding an antigen (higher the affinity of a B cell to recognize antigens, more likely it will clone) Mario Pavone, IBM-KAIST Bio-Computing Research Center p.17/41

15 Clonal Expansion Mario Pavone, IBM-KAIST Bio-Computing Research Center p.18/41

16 Affinity Maturation Affinity is the degree of binding of the cell receptor with the antigen: higher the affinity the stronger the binding and thus better will be the immune recognition and response Each receptor recognizes one antigen, which invades the organism, with different degrees of affinity Affinity Maturation: duringtheimmuneresponse,bcellsandt cells increase the affinity of the cloned receptors The immune response is said to be adaptive because it allows the cells receptors to adapt themselves to antigens, by mutation and selection Clonal selection affect both B cells and T cells, but affinity maturation has been observed in B cells, only Mario Pavone, IBM-KAIST Bio-Computing Research Center p.19/41

17 Binding Shapes Mario Pavone, IBM-KAIST Bio-Computing Research Center p.20/41

18 Somatic Hypermutation Changes to the shape of the receptors are caused by mutations The cloned B cells will undergo a Somatic Hypermutation, creating B cells with mutated receptors Why we use the terminology hypermutation? Hypermutation rate is Inversely Proportional to the cell affinity Such kind of hypermutation, help to preserve high affinity of the cloned cells, and to produce several variants of the receptor selected Thanks to hypermutation, all cloned cells will present slight differences, withrespecttheirparentcells Mario Pavone, IBM-KAIST Bio-Computing Research Center p.21/41

19 The Theory of the Clonal Selection Mario Pavone, IBM-KAIST Bio-Computing Research Center p.22/41

20 Clonal Selection Algorithms (1/2) CSA are inspired by the human s clonal selection principle to produce effective methods for search and optimization Clonal Expansion triggers the growth of a new population of high-value B cells centered on a higher affinity value Hypermutation can be seen as a local search procedure that leads to a fast maturation CSA provides an excellent example of bottom up intelligent strategy [Cutello, Nicosia, Pavone, LNCS 2723, GECCO, 2003] adaptation operates at the local level of cells and molecules, and useful behavior emerges at the global level with the immune humoral and cellular responses. Mario Pavone, IBM-KAIST Bio-Computing Research Center p.23/41

21 Clonal Selection Algorithms (2/2) CSA can be seen as a problem learning and solving system: the Ag is the problem to solve the Ab is the generated solution [Forrest, et al., OxfordUniversityPress,2000] At the beginning of the primary response the antigen-problem is recognized by poor candidate solutions At the end of the primary response the antigen-problem is defeated-solved by good candidate solutions The primary response corresponds to a training phase, whereas the secondary response is the testing phase Mario Pavone, IBM-KAIST Bio-Computing Research Center p.24/41

22 Negative Selection Is the process, which eliminates all T cells, whose receptors are able to recognize and bind with self entities, presentedinthe thymus A blood thymic barrier avoids that nonself entities can be present within the thymus. All cells within it are self entities All T cells, which are not able to recognize any self entities become immunocompetent T cells, i.e.abletoperforman immune response These kinds of cells will be released into the blood stream, with the purpose to patrol the body from the nonself entities This set of T cells, called detectors, candetect any change in self entities or any form of nonself Mario Pavone, IBM-KAIST Bio-Computing Research Center p.25/41

23 Mario Pavone, IBM-KAIST Bio-Computing Research Center p.26/41

24 Mario Pavone, IBM-KAIST Bio-Computing Research Center p.27/41

25 Mario Pavone, IBM-KAIST Bio-Computing Research Center p.28/41

26 How does it works Mario Pavone, IBM-KAIST Bio-Computing Research Center p.29/41

27 Negative Selection Algorithm (1/2) Negative Selection (or negative detection) can be used to perform tasks like pattern recognition by storing information about the set of patterns that are unknown to the system Goal: todetect when elements of a set of self string have changed from an established norm Given a set of self string S, thestandardnegativeselectionalgorithm,isasfollow: [Forrest et al., IEEESymp.onResearchinSecurityandPrivacy,1994] Randomly generate a set of strings, R; Evaluate the match affinity of all strings in R with all strings of S. If the match affinity of a string of R with at least one string of S is greater or equal to a given threshold ε, then it will be eliminated (self-string). Monitor S for changes by continually matching the detectors in R against S The repertoire subset R is known as the detector set Mario Pavone, IBM-KAIST Bio-Computing Research Center p.30/41

28 Negative Selection Algorithm (2/2) Mario Pavone, IBM-KAIST Bio-Computing Research Center p.31/41

29 Negative Selection Approaches (1/3) The first negative selection algorithm was proposed using a binary representation {0, 1} [Forrest et al., IEEESymp.onResearchinSecurityandPrivacy,1994] The r contiguous matching rule was applied to determine the affinity between a detector and an element two elements, with the same length, match if at least r contiguous characters are identical An improved variation of the r contiguous matching rule is the r chunk [Dasgupta, et. al., LNCS2723,GECCO,2003] given an element e and detector d, they match if exist a position p, where all characters of e and d are identical, over a sequence length r Mario Pavone, IBM-KAIST Bio-Computing Research Center p.32/41

30 Negative Selection Approaches (2/3) All matching rules cause undetectable elements (i.e. self elements not seen during the training phase) Crossover closure was proposed with the purpose to find holes each self string will be stored as relations with an attribute foreachbit to reconstruct the original strings one computes the natural join of the relations, producing more than original strings it will return all possible crossovers of the original strings: the total set of strings that are undetectable [Forrest et al., ICARIS,2003] Mario Pavone, IBM-KAIST Bio-Computing Research Center p.33/41

31 Negative Selection Approaches (3/3) Dasgupta et al. (ICARIS 2003), proposed a negative selection algorithm, which operates on a unitary hypercube detector d =(c, r ns ), having a center c and a nonself recognition radius r ns motivation: consider all elements, which are close to the self-center, asself if an element lies within a detectors the it is classified as nonself, otherwiseas self an element e lies within a detector d if the Euclidean distance is smaller than r ns Dasgupta et al. (GECCO 2004), proposed a real-valued negative selection algorithm with variable size detectors (V-detectors) the center of a detector is positioned randomly and must not lie within the hypersphere of a self element Esponda and Forrest (ICARIS 2004), proposed a prototype negative database based on the principles of negative selection ND stores information about the inverse of the data we wish to store. Mario Pavone, IBM-KAIST Bio-Computing Research Center p.34/41

32 Immune Network Theory (1/2) The immune network theory suggest that antibody have portions of their receptors that can be recognized by other antibodies Part of an antibody (paratope) willbindtopartofanantigen (epitope) Also, antibody have epitopes, whichcan be bound by other antibodies paratope Arise a network of communication: immune network The entities presenting bound epitope will be eliminated or repressed, whereastheantibodiespresentingtheactiveparatope will proliferated Such network of stimulatory and suppressive interactions allow a form of associative memory [Hart and Ross, ICARIS, 2002] Mario Pavone, IBM-KAIST Bio-Computing Research Center p.35/41

33 Immune Network Theory (2/2) The immune network algorithm are tightly linked to two equations: matching affinity and change of the antibody s concentration The first models were based on differential equations, which govern the variations in population sizes continuous immune network Immune network are used also as inspiration to the development of machine learning network models with applications in data analysis These kind of models are mainly based on iterative procedures of adaptation discrete immune network Mario Pavone, IBM-KAIST Bio-Computing Research Center p.36/41

34 Immune Network Example Mario Pavone, IBM-KAIST Bio-Computing Research Center p.37/41

35 Danger Theory (DT AIS) (1/3) The central idea in DT-AIS is that the immune system doesn t respond to nonself entity, but to danger Instead to respond to foreignness, the immune system reacts to danger There is no need to attack everything that is foreign Danger Theory is measured by damage to cells indicated by distress signals that are sent out when cells die an unnatural death Cells can die in two ways: apoptotic: normal death that has been requested by the body s internal signaling system necrosis: aformofunexpected death caused by something going wrong with the cell, which often causes an inflammatory response Immune response is contextualized to the location in which necrosis is occurring the danger signal establishes a danger zone around itself Mario Pavone, IBM-KAIST Bio-Computing Research Center p.38/41

36 Danger Theory (DT AIS) (2/3) B cells producing antibodies that match antigens within the danger zone All B cells that don t match or are too far away from danger zone do not get stimulated In the natural immune system is not immediately clear which signals are danger signal the exact nature of the danger signal(s) is still unclear Danger Theory can help to study and understand the intrusion detection systems self-nonself discrimination = danger-nondanger discrimination The concepts of self-nonself may change over time, whereas the ones danger-nondanger are grounded in undesirable events Mario Pavone, IBM-KAIST Bio-Computing Research Center p.39/41

37 Danger Theory (DT AIS) (3/3) Instead of responding directly to a nonself entity, the immune system responds to cells, which are under stress? [Matzinger, Science, 2002] Mario Pavone, IBM-KAIST Bio-Computing Research Center p.40/41

38 References D. Dasgupta, Artificial Immune Systems and Their Applications, Springer-Verlag, 1999 L. N. de Castro and J. Timmis, Artificial Immune Systems: A New Computational Intelligence Approach, Springer-Verlag,2002 A. O. Tarakanov, V. A.Skrormin and S. P. Sokolova, Immunocomputing: Principles and Appilcations, Springer-Verlag,2003 International Conference on Artificial Immune Systems (ICARIS), 2002, 2003, 2004, Mario Pavone, IBM-KAIST Bio-Computing Research Center p.41/41