Molecular genetic programming

The paper addresses a new implementation of genetic programming by using molecular approach. Our method is based on data¯ow techniques in DNA computing. After description of fundamental operations on DNA molecules and construction of logical functions the genetic programming method is introduced. We propose a way to handle these graph encoding molecules and which can be considered a genetic programming algorithm; a short discussion about experiments in implementing parts of this procedure is added. 1 Introduction Genetic programming [1, 2] has been recently developed as one of evolutionary algorithms. Their earlier implementations are called: genetic algorithms [3] and evolution strategies. On the whole, genetic programming is a methodology to solve problems by genetically breeding populations of computer programs. In such an approach for a particular problem sets of functions and terminals are created. An initial population of LISP-like expressions is a collection of random tree-like or graph-like compositions of fundamental functions and terminals. Each expression called also a program represents a possible solution to the problem and is evaluated against this problem. Genetic operators of selection and crossover are applied to create new populations of programs. Evolutionary process is continued until either a solution is found or a maximum number of generations is reached. Thus, while terminate condition is not true, the following cycle is performed: increase the number of generation; select population(t) from population(t À 1); recombine population(t), using crossover and mutation operators; evaluate population(t). With the help of genetic programming, genetic operations on graph-like structures are performed. These structures can describe e.g. logical functions. Genetic operators like selection and crossover create new populations of graphs in order to ®nd the most appropriate solutions. However, in traditional computers such programs are not performed in parallel yet. In the von Neumann machines instructions are written in the memory one after another except for those after instructions of jumps to the other memory parts. Instruction pointer aims usually at the ready to sending (to a computational unit ± processor), next instruction. Multitasking, sometimes real-time and with parallel execution systems e.g. Unix and Linux use the very quick task exchange in the computational unit. Such unit can execute only one task. More privileged tasks are performed more often. Even after adding several thousands of processors there are still great problems with fully parallel execution of tasks. And some tasks are still performed in the sequences on appropriate computational units. On the other …