Related terms:
- Artificial Immune Systems
- Classification (Machine Learning)
- Genetic Algorithm
- Cell Population
- Clonal Selection
- Immune Algorithm
When Evolutionary Computing Meets Astro- and Geoinformatics
Zaineb Chelly Dagdia Dr, Miroslav Mirchev Dr, in Knowledge Discovery in Big Data from Astronomy and Earth Observation, 2020
Clonal Selection Theory
Clonal selection theory (Burnet et al., 1959) is used to clarify the basic response of the adaptive immune system to antigenic stimuli. Clonal selection involves two main concepts i.e., are cloning and affinity maturation. More precisely, it establishes the idea that only those cells capable of recognizing an antigen will proliferate, while other cells are selected against. Clonal selection calls both B and T cells. When B cell antibodies bind with an antigen, cells become activated and differentiated either to be plasma cells or memory cells. The closer the matching between an antibody and a specific antigen is, the stronger is the bond. This property is called affinity. Plasma cells make large amounts of a specific antibody that work against a specific antigen to destroy it. Memory cells remain with the host and promote a rapid secondary response. However, before this process, clones of B cells are produced and undergo somatic hypermutation. Consequently, diversity is introduced into the B cell population. Moreover, a selection pressure is performed, which implies a survival of the cells with higher affinity. Let us note that clonal selection is a kind of an evolutionary process, where an antibody represents a solution, the affinity defines the fitness function, and the antigen represents the value of the objective function to optimize. The cloning process is seen as the reproduction of solutions, the somatic hypermutation represents the mutation of a solution, and the affinity maturation represents the mutation and the selection of best solutions. Based on this theory, various clonal selection algorithms have been proposed in the literature, and most of them are devoted to optimization problems. A detailed description and comparison of AIS clonal selection algorithms can be found in Ulutas and Kulturel-Konak (2011).
View chapterPurchase book
Read full chapter
URL:
https://www.sciencedirect.com/science/article/pii/B9780128191545000266
Clonal-Selection-Based Minimum-Interference Channel Assignment Algorithms for Multiradio Wireless Mesh Networks
Su Wei Tan, ... Cheong Loong Chan, in Bio-Inspired Computation in Telecommunications, 2015
13.3.1.3 Clonal selection and expansion
In this step, all B-cells in the current population P are subject to all or some of the following operations:
- –
Cloning. Each B-cell in P is cloned multiple times to produce a clone population.
- –
Affinity maturation. Each of the clones undergoes a hypermutation process.
- –
Aging. In this operation, all B-cells exceeding a maximum age will be deleted from the population. The age of a B-cell is increased at each generation.
- –
Metadynamics. This models the ability of the immune system in continuous production and recruitment of novel structures. Typically, some randomly generated B-cells are added to the population.
CLONALG, BCA, and CLICA each adopts slightly different operations, as described below
13.3.1.3.1 CLONALG
In CLONALG, the following operations are executed in order:
- i.
Cloningii
- ii.
Affinity maturationiii
- iii.
Metadynamics
(Video) Somatic Hypermutation And Affinity Maturation
Cloning. Following the optimization version of CLONALG (de Castro and Von Zuben, 2002), each B-cell will be cloned round(β·N) times. This results in a total of Nc clones for population P,
(13.5)
where β is a configurable parameter, N is the total number of B-cells, and round() is the operator that rounds its argument toward the closest integer.
Affinity maturation. In this step, each of the clones undergoes a hypermutation process. The number of mutations M is determined by a mutation potential. In Tan (2010), we adopted the mutation potential used in CLONALG, which calculates the mutation rate α as
(13.6)
where α is a configurable parameter and f is the fitness function value normalized in [0,1]. Cutello et al. (2005a) compared the original CLONALG (uses Equation (13.6)) with a variant that uses the following mutation potential:
(13.7)
Originally proposed in de Castro and Timmis (2002). They tested both versions with four different classes of optimization and pattern recognition problems, and showed that the setting of the mutation rate and the parameter ρ is crucial for the algorithm performance. Both Equations (13.6) and (13.7) are considered in this chapter, and are referred to as hyper1 and hyper2, respectively. It can be seen that the mutation rate α is inversely proportional to the affinity; i.e., the higher the affinity, the smaller the value of α, and vice versa. This is a distinct characteristic of hypermutation in both equations.
The mutation is done by a simple random mutation operating on the vertex visiting order. In particular, the mutation operator randomly chooses M times two positions of a clone, and their respective elements are swapped. Given L as the length of the clone receptor and the affinity f of a clone, M is given by
(13.8)
The above mutation operator is referred to as m-point random-swap, as multiple swaps are performed on each B-cell. In this chapter, we also investigate a simpler mutation operator, which allows only one swap for each B-cell. This is referred to as 1-point random-swap. This operator loops through each position of a B-cell to find the first swap position with probability α. Once a position is found, the loop is ended and swapping is done with a randomly chosen target.
Following the mutation operation, the mutated clones are added to population P. After that, B-cells in P undergo an elitist selection in which the best N B-cells are retained.
Metadynamics. Following CLONALG, the d lowest affinity B-cells are replaced by some newly created random solutions.
13.3.1.3.2 BCA
In BCA, the following operations are executed in order:
- i.
Cloning
- ii.
Metadynamics
- iii.
Affinity maturation
Cloning. In BCA (Kelsey et al., 2003b), each B-cell x is cloned to produce a clonal pool, Cx, and all adaptation takes place within Cx. Similar to CLONALG, the size of Cx can be given by round(β·N). In the literature, β=1 is typically considered.
Metadynamics. In order to maintain diversity within the search, a clone is selected at random from Cx, and elements in the chosen clone will be randomized.
Affinity maturation. BCA introduced a novel mutation operator, which operates on contiguous regions of a vector. Applied to function optimization problems where each B-cell is represented as a vector of bit string of 64 bits, the operator randomly chooses a site (called hotspot) within the vector, along with a random length. Each element within the contiguous region is subject to individual mutation. The motivation behind this operator is to offer a more focused search.
We adapt the contiguous mutation operator to our problem as follows. First, a random site and a random length are chosen. Then, elements within the contiguous region are randomized. Note that the operator does not consider the possible influence of the affinity. To take this into consideration, we propose to calculate the length of the contiguous region by Equation (13.8). This results in two simple variants that use Equations (13.6) and (13.7) to calculate the mutation rate α, respectively. The above three operators are referred to as contiguous-random, contiguous-hyper1, and contiguous-hyper2, respectively.
Following the mutation operation, the mutated clones are compared to their respective parents. A parent B-cell will be replaced if there exists a clone with a better affinity.
13.3.1.3.3 CLIGA
In CLIGA, the following operations are executed in order:
- i.
Cloning
- ii.
Affinity maturation
- iii.
Aging
Cloning and affinity maturation. Similar to CLONALG, CLIGA generates for each B-cell round(β·N) number of clones. Unlike CLONALG, which performs cloning for all B-cells in P, cloning is done stochastically in CLIGA for each B-cell based on a cloning potential. In Cutello et al. (2003), a truncated exponential V(f(x))=e−k(l−f(x)) is used, where f(x) is the affinity value of B-cell x, k is a configurable parameter, and l is the length of the vector representing a B-cell. In the affinity maturation step, a straight-line mutation potential M(f(x))=1−(l/f(x)) is considered.
In this chapter, CLIGA is considered mainly to study the impacts of aging on the algorithm performance. Hence, we adopt the approach taken by CLONALG for both cloning and affinity maturation in our implementation of CLIGA. Specifically, static cloning is used instead of stochastic cloning, and mutation potential Equation (13.6) is used.
Aging. In this step, old B-cells will be eliminated. When the aging operation is in place, each B-cell is assigned an age, which is increased on each new generation. In addition, a newly created clone inherits the age of its parent, and its age is reset if it is successfully mutated. In previous work (Cutello et al., 2003, 2005a,b; Cutello and Nicosia, 2005), two classes of aging operators have been considered: (1) static and (2) stochastic. In static aging, B-cells with ages exceeding a maximum age τ will be eliminated; in stochastic aging, each B-cell is subjected to elimination with a probability Pdie(τ)=(1−e(−ln(2)/τ)). Within each class, two approaches are possible: (1) pure aging and (2) elitist aging. In elitist aging, when a new population for the next generation is generated, we do not allow the elimination of B-cells with the best affinity. On the other hand, pure aging does not impose such restrictions. Consequently, there are four variants of aging operators in total:
- 1.
Static-pure aging
- 2.
Static-elitist aging
(Video) Antibody affinity maturation- 3.
Stochastic-pure aging
- 4.
Stochastic-elitist aging
We experiment with each of the variants in our evaluations.
At the end of the aging operation, if only (N′<N) B-cells survived, the (μ+λ)-selection operator is used to create (N−N′) B-cells to maintain a constant population size. In the original CLIGA, a stopping criterion based on information gain is used. This is not included in our implementation of CLIGA, so as to have the same stopping criterion (i.e., number of generations) used for our CLONALG and BCA implementations.
View chapterPurchase book
Read full chapter
URL:
https://www.sciencedirect.com/science/article/pii/B9780128015384000136
Volume 1
R. Boyton, ... D.M. Altmann, in Encyclopedia of Respiratory Medicine(Second Edition), 2022
Adaptive Immunity
The specific, adaptive immune response to infection in the lung depends on the specific recognition of foreign, microbial antigens by the clonotypic receptors of B and T lymphocytes. The antibodies against a given pathogen may be directed against several different binding sites, termed epitopes. Some antibodies are able to neutralize infections through interference with a vital, structural component of the microbe. Others may clear infections through opsonization for phagocytosis and through activation of the complement system. After initial exposure, a state of B cell memory is established, leading to the ability to mount a more rapid IgG antibody response on any subsequent re-exposure. Furthermore, the B-lymphocyte-derived IgG response is unique in the immune response in showing affinity maturation. This is the ability to introduce small numbers of random mutations into the genes for the antibody, somatically mutating the sequence so that receptors with better affinity for the epitope are selected at each generation of cell division. This antibody response and its affinity maturation are absolutely dependent on cytokine-mediated help ensuing from antigen recognition by CD4 Tcells.
The antibody response to infection mounted by B lymphocytes acts alongside the T lymphocyte response. T lymphocytes carry the surface markers CD3 along with either CD4 or CD8, mature in the thymus, and express an antigen receptor that must recognize peptide fragments of microbial antigen presented by host HLA molecules. T lymphocytes mediate their effector function either through the release of a wide range of soluble mediators called cytokines or, in the case of CD8 cytotoxic Tcells, by direct killing of target cells. Early studies aimed at cloning the genes for the antigen receptor used by T lymphocytes identified two separate systems of receptors now known to delineate separate lineages of thymus-differentiated Tcells. One receptor heterodimer was the αβ receptor, the other the γδ receptor. The cells expressing the αβ Tcell receptor are the conventional CD4 and CD8 populations with the properties of mounting cytolytic responses to infected cells, making cytokines, and stimulating B cells. This involves recognition of peptide fragments of micorobial antigens loaded through the biosynthetic pathway of classical HLA class I or class II molecules into the peptide-binding groove. The Tcells expressing the γδ receptor are particularly associated with defense at epithelial surfaces, have a different mode of recognizing foreign antigens, but are also involved in lung immunity. They do not need antigen processing or major histocompatibility complex (MHC) restriction for antigen recognition. The types of antigen they recognize can be categorized into those that are widely distributed and are constitutively expressed in host cells and by microbial pathogens, and those that are inducibly expressed and might be restricted to specific cell types. Presentation may be by classical MHC molecules or by related molecules such as CD1d. Examples of the families of recognized antigens include the nonprotein pyrophosphates and alkylamines that are found in bacteria, plant or animal cells, and bacterial and mammalian homologs of the stress protein Hsp60/GroEL.
CD4-positive, TCR αβ expressing T lymphocytes are termed Th cells. Their effector function is through the release of cytokines that stimulate or activate some other cell type: for example, IL-4, which stimulates B cells and attracts eosinophils; IL-2, which drives other T lymphocytes, including CD8 cells to divide; and IFN-γ, which causes other T lymphocytes to differentiate at the same time as it activates macrophages and dendritic cells. CD4 cells can themselves kill target cells through the release of TNF-α.
CD8-positive, TCR αβ T lymphocytes are termed cytotoxic cells (or sometimes, killer cells). On recognition of microbial peptides presented by HLA class I molecules, they function by lysing infected cells, particularly by release of granules of perforin.
View chapterPurchase book
Read full chapter
URL:
https://www.sciencedirect.com/science/article/pii/B9780081027233002675
Calculating germinal centre reactions
Lisa Buchauer, Hedda Wardemann, in Current Opinion in Systems Biology, 2019
Conclusions
Computational models of affinity maturation can be used to explore GC mechanisms as well as vaccination strategies. Interesting findings often emerge if such models are developed in conjunction with experimental studies, as mathematical models can assist the interpretation of experimental data and the development of novel testable hypotheses. Half of the studies reviewed here are directly available in online repositories (Table1). This facilitates re-use and adaptation of existing models. In the future, computational models may mature to allow prediction of vaccination outcomes based on key parameters such as vaccine composition, administration schedule, and the pre-vaccination B-cell repertoire of individual patients.
View article
Read full article
URL:
https://www.sciencedirect.com/science/article/pii/S2452310019300447
The use of computational intelligence in intrusion detection systems: A review
Shelly Xiaonan Wu, Wolfgang Banzhaf, in Applied Soft Computing, 2010
The clonal selection algorithm implementing affinity maturation is now gradually developed into a new computational paradigm. CLONALG (CLONal selection ALGorithm) [75], ARIS (Artificial Immune Recognition System) [278], and opt-aiNet [72] are well known clonal selection algorithms. These algorithms are used in performing machine-learning and pattern recognition tasks, and solving optimization problems. Although they employ the generation-based model and evolutionary operators when generating offspring, they distinguish themselves from other evolutionary algorithms by the following: firstly, cloning and mutation rates are decided by an individual’s affinity. The cloning rate is proportional to the affinity, while the mutation rate is inversely proportional to the affinity. There is no crossover in clonal selection algorithms; secondly, it is a multi-modal preserving algorithm. The memory cell population () incrementally saves the best solution in each generation. will be returned as the final solution when the algorithm is terminated; thirdly, the population size is dynamically adjustable. Applications of these algorithms to intrusion detection can be found in [123,204,205,283]
View article
Read full article
URL:
https://www.sciencedirect.com/science/article/pii/S1568494609000908
Computational intelligence for preventive maintenance of power transformers
Shen Yuong Wong, ... Hui Hwang Goh, in Applied Soft Computing, 2022
4.10 Artificial Immune System (AIS)
An Artificial Immune System (AIS) is a type of intelligent search-based model constructed on the basis of the immune network theory, clonal selection principle, and affinity maturation principle[150]. In general, an AIS comprises a network of nodes, which represent the antibodies connected by edges to form the node pairs. The antibodies represent the internal images of input patterns, which are also referred to as antigens in the immune theory. In an AIS, the clonal selection theory is utilized to adapt the network of antibodies to train the antigen patterns, during which a network of memory cells is built to determine the existence of data clusters. Through training, the Euclidean distance is used to measure the affinity between the training pattern and antibody. The antibody with the highest affinity is selected to be cloned. Meanwhile, the mutation process is included in the clone process for improving the antigens recognition. The antibody with the highest affinity mutated clones is selected as the memory set. Referring to[151], an AIS is useful for global searching. Its capability in avoiding premature convergence to local optimal solutions is also outstanding.
In 2014, Malabanan etal.[150] presented an algorithm by hybridizing an AIS with the Radial Basis Function Neural Network (RBFNN) to process and analyze the DGA data for transformer fault diagnosis. The DGA data set is firstly provided to the AIS as the antigens. The resulting memory cells of antibodies are used as the centers of the hidden layer nodes of the RBFNN. Two DGA data sets are used for training and test, namely set A (200 samples) and set B (150 samples), respectively. For comparison, the k-means algorithm is used to replace the AIS to be combined with the RBFNN for the diagnostic task. The test accuracy rates of AIS-RBFNN are 76.16% and 83.17% for set A and set B, while those of k-means RBFNN are 74.80% and 80.41%, respectively.
In the same year, a hybridized AIS Fuzzy C-mean Clustering model was implemented by Wang etal. for DGA analysis[151]. The DGA data set is firstly processed by the AIS. Subsequently, the memory sets composed by the identified gas characteristics are used for training the fuzzy c-mean model, in which the transformer fault diagnosis is conducted. In this research, a total of 361 groups of DGA data are collected, in which 80 groups are randomly selected from them as the training antigens. The proposed model produces diagnostic accuracy rates between 84.3% and 91.6% with respect to thermal faults and between 85.3% and 95.6% with respect to discharging faults.
For analyzing the internal insulation faults of power transformers, Li etal. combined both AIS and PSO with the RBF neural networks to analyze the dissolved gas data[152]. In this research, the fault samples serve as the input to an AIS. The derived results of the AIS are utilized to identify the location of the initial centers of the hidden layer nodes of the RBFNN. Additionally, the network weights are optimized by the PSO algorithm, A total of 120 groups of DGA data are used for model training, while another 180 groups of data are used for test. For comparison, the BPNN and single RBFNN models are applied to the collected data, and their diagnostic accuracy rates are 83.4% and 78.3%, respectively. On the other hand, the diagnostic accuracy rate of AIS-PSO-RBFNN is 91.7%, indicating its effectiveness for DGA-based transformer fault diagnosis.
Based on[150,151], the AIS has great capability in data pre-processing and analysis, especially for DGA data sets. The diagnostic results in[151,152] indicate that AIS-based hybrid models can achieve over 90% accuracy. A comparison between both the AIS and k-mean algorithm in[150] also demonstrates the usefulness of the AIS. Nevertheless, as mentioned in[151], the AIS has no clear clustering centers, and it is rarely used as single model for transformer fault analysis. In other words, the AIS is usually combined with other methods in order to improve the diagnostic performance. In addition, the AIS shows a weak performance in learning the same fault types since it lacks fault-type information in the antigen schema[153].
View article
Read full article
URL:
https://www.sciencedirect.com/science/article/pii/S156849462101005X
Host-pathogen interactions and the immune response to infectious diseases
Rodrigo Abreu, Ted M. Ross, in Current Opinion in Systems Biology, 2018
The adaptive immune response
Protective adaptive immunity to influenza is generally associated with a strong humoral response against virus surface proteins HA and NA [18]. The specific humoral response to influenza is driven by high affinity antibodies that can derive from multiple immune sources [29–31]. Long-lived plasma cells resulting from past influenza infections constitutively secrete antibodies and are responsible for the pre-immunity frequently observed in influenza vaccination trials (further discussed below).
During primary exposure to influenza or in the absence of effective constitutive humoral memory, IgM+IgD+ naïve B cells with low to medium affinity promiscuous B-cell receptor (BCR) are activated by influenza antigens in primary follicles of the peripheral lymphoid organs [32]. Activated B-cells migrate to the intrafollicular zone where they interact with activated CD4 follicular T helper cells, proliferate and isotype-switch. Some of these expanded B-cells become low-affinity memory B cells or differentiate into short-lived plasma cells actively secreting non-affinity maturated antibodies. Meanwhile other isotype-switched activated B-cells migrate to the germinal center (GC) for somatic hypermutation, clonal selection and affinity maturation. Positively selected clones migrate out of the GC to differentiate into long-lived plasma cells or become high-affinity GC-dependent memory B-cells [29,32]. During secondary exposure, either by infection or recurrent vaccination, high affinity GC-dependent and low affinity GC-independent memory B cells rapidly expand and differentiate into long-lived or short-lived plasma cells or migrate back to GC for further affinity maturation [33]. In addition to the classical isotype-switched memory B-cells, antigen-driven T-cell-independent activated B-cells can also acquire a long-lived memory phenotype during intraperitoneal bacterial infection [34] (Figure2). However, the relevance of T-cell-independent B-cells during influenza infection or other respiratory viral infections remains to be experimentally addressed.
Figure2. Adaptive humoral response to influenza virus infection. Antigen-activated dendritic cells travel to the lymph node and activate naive CD4 T cells in the T-cell zone (a), while soluble antigens activate naïve B cells in the B-cell follicle through direct BCR signaling (b). Low-affinity IgM mature B cells can exit the B cell follicle and differentiate into plasma cells independent of T-helper cells expansion (c). Most influenza-specific activated B cells migrate to the periphery of the B-cell follicle and interreact with influenza specific CD4 T follicular helper cells (Tfh) for proliferation and isotype switch (d). Low-affinity IgG or IgA influenza-specific B cells exit the germinal center, differentiate into short-lived plasma cells or memory B cells (Bmem) and travel to the site of infection (e). Meanwhile, some influenza-specific B cells migrate to the germinal center for clonal expansion and affinity maturation (f). High affinity mature B cells migrate out of the germinal center and differentiate into long-lived plasma cells in the bone marrow, or high affinity memory B cells that will reside at the site of infection (g).
Adaptive cell-mediated immunity also plays an important role for protection against influenza infection particularly in heterosubtypic challenge models [35,36]. As mentioned above CD4 T-cells are crucial for memory B-cell development. Furthermore, influenza-specific memory CD4 T-cells promote a protective cytokine milieu that favors viral clearance with low immunopathology and synergize with naïve CD8 T-cells to promote cytotoxic activity against influenza infected cells [37] (Figure3). Unlike the humoral immune response, which is predominantly directed against highly variable HA and NA surface proteins, influenza-specific T-cells can recognize internal conserved antigens as result of cross-presentation by dendritic cells and macrophages [38]. Particularly after infection, Richards and colleagues [39] found significant expansion of influenza-specific CD4 T cells in mice. After primary infection of naïve animals with pandemic human influenza virus H1N1, these cells recognize linear epitopes from both surface proteins such as HA, as well as internal antigens (NS and NP) present in both homologous and heterologous strains. Moreover, these memory CD4 T cells retain the antigenic breath observed at peak of infection, with a slight enrichment towards HA reactive cells [39]. Similar results were observed in human cohorts where PBMCs from healthy donors had significant numbers of H1, H3, N1, N2 and NP and matrix reactive CD4 T cells, mostly likely derived from past infections [40] because vaccination with current inactivated vaccines has minimal to no impact on CD4 T-cell expansion. Altogether, these studies show that despite not conferring sterilizing immunity, memory cellular immunity can ameliorate disease against antigenically drifted strains and heterosubtypic infection [41].
Figure3. Adaptive immune responses to secondary influenza virus infection. Secondary influenza infection is inhibited by IgA neutralizing antibodies at the mucosal surface that prevent viral adhesion and internalization (a). Drifted virus strains can evade neutralizing antibodies, but might still be recognized by broadly reactive non-neutralizing IgG antibodies (b). Antibody-opsonized viral particles are quickly phagocytized by innate immune cells such as dendritic cells (DCs) and macrophages (Mφ), and induce proinflammatory cytokine secretion (c). Activated DC and Mφ will present processed antigens to tissue resident CD4 memory T cells from previous infections (d). Meanwhile recruited memory B cells will be activated by soluble antigens, proliferate and differentiated in antibody-secreting plasma cells (e). Activated memory T-cells secrete high amounts of IFN?? and drive lymphocyte recruitment (NK and CD8 T cells) important for clearance of influenza infected cells (f).
View article
Read full article
URL:
https://www.sciencedirect.com/science/article/pii/S2452310018300866
The Impact of Soft Computing for the Progress of Artificial Intelligence
Dipankar Dasgupta, ... Fernando Nino, in Applied Soft Computing, 2011
2.1.3 Clonal selection algorithm
Clonal selection theory states that a clonal expansion of the original lymphocyte occurs when the original lymphocyte is activated by binding to the antigen; however, any clone of the activated lymphocyte with antigen receptors specific to molecules of the organism's own body (self-reactive receptors) is eliminated during the development of the lymphocyte. During the clonal expansion of B cells, the average affinity increases for the antigen that triggered the clonal expansion through a process of affinity maturation. Therefore, the memory B cells are developed to make a more effective immune response to antigens that had been encountered. Affinity maturation is caused by a somatic hypermutation and a selection mechanism. Somatic hypermutation results in the diversity of antibodies by introducing random changes to the genes that encode for them. The selection mechanism guarantees that only those clones (antibodies) with higher affinity for the encountered antigen will survive. On the basis of clonal selection theory, clonal selection algorithm was initially introduced in [11] and formally described in [12]. The general algorithm was named CLONALG.
Work by Ciccazzo et al. [64] introduced a new clonal selection algorithm called the elitist Immune Programming (EIP). EIP was an extension of Immune Programming (IP) and the pseudo-code of EIP was provided. Firstly, the concept of elitism borrowed from other immune inspired algorithm is introduced to EIP; that is, at each generation g, the best solution found so far cannot be erased from the population. In addition, EIP uses a new class of hypermutation operators and a network-based coding. A hypermutation operator acts on one component, link or node at a time. All the operators take or return only consistent circuits. This work introduced ten ad-hoc network-based hypermutation operators, including add-series, add-parallel, add-random-component, expand-node, delete-component, mutate-component-value, copy-component-value, mutate-component-kind, link-modify, and shrink. This work applied the EIP algorithm to the synthesis of topology and sizing of analog electrical circuits. The quality of the circuit was assessed by the distance between the curve described by a circuit and the one described by a hypothetical ideal circuit. The experiments showed that the circuits obtained by EIP were better than the one found by Genetic Programming.
Halavati et al. [65] added the idea of symbiosis to CLONALG. As a variation of CLONALG, this algorithm uses partially specified antibodies that may not have all required data to be evaluated as a solution. This algorithm is initialized with a set of partially specified antibodies, each having just one specified property. Then, the algorithm randomly picks an antibody to add to an assembly. By repeatedly doing so, the algorithm shall build an assembly with all required properties. If the algorithm fails to complete such assembly with existing antibodies, it will create some antibodies with random values for all missing positions of the created assembly and add them to the population. The process of cloning, mutation and selection in this algorithm is extremely similar to those in CLONALG. The inspiration of using partially specified antibodies comes from the assumption that a problem can be broken into several sub-problems and thus good solutions of these sub-problems may compose a general good solution for the main problem. If such assumption becomes true, this algorithm may find a solution faster than CLONALG. By applying them to multimodal function optimization and combinatorial optimization, this work showed that the proposed algorithm can solve problems that CLONALG failed to solve.
The work in [66] presented a variation of CLONALG for software mutation testing. The variation allowed several memory individuals to contribute to the recognition of an antigen, rather than following the concept of a memory individual per antigen in CLONALG. Each antibody represents a single test in this application. The population of antibodies is initialized with s tests, either by randomly generated tests or pre-specified ones. The algorithm iteratively searches for those antibodies that will kill at least one mutant program that has not already been killed by an existing memory test. The mutation score (MS) is used to evaluate the affinity (or fitness) of an antibody. Those antibodies with high affinity are added to the memory set to be returned to the tester at the end of the process. The process of antibody evolution (clone selection) in this variation is same as that in CLONALG. The effectiveness of the proposed method was compared against an elitist genetic algorithm and the results showed that the proposed method generates higher mutation scoring test sets with less computational expense.
Inspired by the clonal selection principle, work in [67] developed an immune algorithm for the protein structure prediction problem on lattice models. Antigens (Ags) and B cells in the proposed immune algorithm represent a sequence of hydrophobic-pattern of the given protein s∈{H, P}ℓ and a sequence of directions r∈{F, L, R}ℓ−1, respectively. H, P denotes two types of beads in the standard Dill's lattice model: bead-Hydrophobic/non-polar or bead-hydrophilic/Polar, respectively. F, L, R denotes Forward, Left, and Right, respectively. ℓ is the number of amino acid in the protein sequence. Different from the original clonal selection algorithm, this work proposed two special mutation operators (inversely proportional hypermutation and hypermacromutation) and one aging mechanism. Inversely proportional hypermutation is similar to the mutation operator in the conventional clonal selection algorithm. It makes mutations inversely proportional to the fitness value. The hypermacromutation operator randomly determines the number of mutations, which does not use functions depending on constant parameters. The aging operator is designed to avoid getting trapped in local minimum by eliminating old B cells from the population based on the maximum number of generations allowed for generated B cells to be remained in the population. This work also discussed the characteristic dynamics of the proposed immune algorithm and demonstrated the competitive performance of the algorithm by testing with well-known protein structure prediction lattice models, the HP model in two-dimensional and three-dimensional square lattices, and the functional model protein.
Wilson et al. [68] proposed Trend Evaluation Algorithm (TEA) to evaluate price time series data. TEA is very similar to CLONALG with some differences that are summarized in this work. TEA not only finds the best fitting candidates (long term memory pool) but also maintains short term memory pool by proliferating all bound trackers. Apoptosis and mutation in the TEA occur across all population members. A simple antigen A containing 20 fictitious price movements and 8 trends (T1–T8) is constructed to test the ability of the TEA to identify price trends. Antigen A is further splitted into two subsets A1 an A2. A1 contains three simple trends T1, T2, and T3 and the other more complex trends are involved in A2. Four experiments were defined in this work to examine the algorithm's ability to identify the price trends and to investigate the influence of the long term memory pool.
View article
Read full article
URL:
https://www.sciencedirect.com/science/article/pii/S1568494610002723