Drug Design , Artificial Intelligence Methods in

Ant colony optimization Ant colony optimization 15 (ACO) is an agent-based algorithm procedure in16 spired by the function of ant colonies and their search 17 for the optimum path to food sources. The virtual 18 agents are called artificial ants or ants, and the opti19 mization problem is represented as a trail-and-error 20 search for the optimum path on a weighted graph. The 21 pheromone that is deposited by ants on the trail is 22 represented as weights for graph components (vertices 23 or edges). Each ant generates a solution by moving 24 on the graph and by selecting the next step based 25 on the pheromone level. The pheromone level is up26 dated after each cycle (when all ants found a solution) 27 by adding a pheromone quantity proportional to the 28 quality of the solutions to which it belongs. 29 Antigen An antigen is a molecule (chemical compound, 30 protein or polysaccharide) that induces an immune re31 sponse. Each pathogen contains specific antigens that 32 are recognized by the immune system. The antigen re33 gion that is recognized by the immune system is called 34 an epitope. 35 Antibody An antibody (or immunoglobulin) is a pro36 tein used by the immune system to identify bacteria, 37 viruses and other pathogens or foreign molecules. The 38 antibody region that binds antigens is extremely vari39 able, thus allowing the immune system to recognize 40 a large diversity of pathogens. The ability to recognize 41 antigens is improved through successive cycles of anti42 gen presentation, antibody cloning, and hypermuta43 tion of the variable region of the antibody. 44 Artificial immune systems Artificial immune systems 45 (AIS) represent a class of optimization algorithms 46 inspired by the components and mechanisms of the 47 biological immune system. AIS simulate the learning 48 and memory capabilities of the immune system to 49 develop computational algorithms for pattern recog50 nition, function optimization, classification, process 51 control, and intrusion detection. 52 Genetic algorithms Genetic algorithms (GA) solve high53 dimensional problems through a Darwinian evolution 54 of a population of individuals, in which each individual 55 (chromosome) represents a possible solution. Depend56 ing on the type of the optimization problem, chromo57 somes may represent the solution in a binary, continu58 ous, or hybrid encoding. Each chromosome has a fit59 ness value that measures the quality of the solution. 60 A population of parents evolves to a generation of chil61 dren by crossover and mutation. 62 Particle swarm optimization Swarm intelligence (SI) 63 represent a group of distributed intelligence algo64 rithms that solve optimization problems by applying 65 processes inspired by swarming, herding, and flocking 66 of various species. Particle swarm optimization (PSO) 67 simulates the swarming behaviors observed in swarms 68 of bees, flocks of birds, or schools of fish. PSO con69 siders a swarm of particles that start from a random 70 position and have a random velocity. At each step 71 a particle moves to a new position that is determined 72 by its own experience (the best past position) and by 73 thememory of the best particle in the swarm. PSOmay 74 be applied to both binary and continuous optimization 75 problems, and its main strength is a fast convergence. 76 Quantitative structure-activity relationships 77 Quantitative structure-activity relationships (QSAR) 78 represent regression models that define quantita79 tive correlations between the chemical structure of 80 molecules and their physical properties (boiling point, 81 melting point, aqueous solubility), chemical properties 82 and reactivities (chromatographic retention, reaction 83 rate), or biological activities (cell growth inhibition, 84 enzyme inhibition, lethal dose). The fundamental 85 hypotheses of QSAR is that similar chemicals have 86 similar properties, and small structural changes result 87 in small changes in property values. The general form 88 of a QSAR equation is P(i) D f (SDi ), where P(i) is 89 a physical, chemical, or biological property of com90 pound i, SDi is a vector of structural descriptors of i, 91 and f is a mathematical function such as linear regres92 sion, partial least squares, artificial neural networks, or 93 support vector machines. A QSAR model for a prop94 erty P is based on a dataset of chemical compounds 95