E. Cantú-Paz et al. (Eds.): GECCO 2003, LNCS 2723, pp. 413–424, 2003. © Springer-Verlag Berlin Heidelberg 2003 DNA-Like Genomes for Evolution in silico Michael West, Max H. Garzon, and Derrel Blain Computer Science, University of Memphis 373 Dunn Hall, Memphis, TN 38152 {mrwest1, mgarzon}@memphis.edu, derrelb@earthlink.net Abstract. We explore the advantages of DNA-like genomes for evolutionary computation in silico. Coupled with simulations of chemical reactions, these ge- nomes offer greater efficiency, reliability, scalability, new computationally fea- sible fitness functions, and more dynamic evolutionary algorithms. The proto- type application is the decision problem of HPP (the Hamiltonian Path Prob- lem.) Other applications include pre-processing of protocols for biomolecular computing and novel fitness functions for evolution in silico. 1 Introduction The advantages of using DNA molecules for advances in computing, known as bio- molecular computing (BMC), have been widely discussed [1], [3]. They range from increasing speed by using massively parallel computations to the potential storage of huge amounts of data fitting into minuscule spaces. Evolutionary algorithms have been used to find word designs to implement computational protocols [4]. More recently, driven by efficiency and reliability considerations, the ideas of BMC have been ex- plored for computation in silico by using computational analogs of DNA and RNA molecules [5]. In this paper, a further step with this idea is taken by exploring the use of DNA-like genomes and online fitness for evolutionary computation. The idea of using sexually split genomes (based on pair attraction) has hardly been explored in evolutionary computation and genetic algorithms. Overwhelming evidence from biology shows that “the [evolutionary] essence of sex is Mendelian recombina- tion” [11]. DNA is the basic genomic representation of virtually all life forms on earth. The closest approach of this type is the DNA-based computing approach of Adleman [1]. We show that an interesting and intriguing interplay can exist between the ideas of biomolecular-based and silicon-based computation. By enriching Adle- man’s solution to the Hamiltonian Path Problem (HPP) with fitness-based selection in a population of potential solutions, we show how these algorithms can exploit bio- molecular and traditional computing techniques for improving solutions to HPP on conventional computers. Furthermore, it is conceivable that these fitness functions may be implemented in vitro in the future, and so improve the efficiency and reliabil- ity of solutions to HPP with biomolecules as well.