Comparison of Simulated Annealing and Genetic Algorithm Approaches in Optimizing the Output of Biological Pathways

Biological processes are random in nature and studying them through laboratory experiments alone is costly and time-consuming. Applying mathematical modeling provides a more efficient method for understanding and modifying them. Recently we proposed an algorithm Box to assist in modeling and controlling any biological pathway. In the Box algorithm, the output (protein concentration) control and optimization is carried out by applying simulated annealing to selective high-sensitivity reactions. Here we apply an alternative approach, a genetic algorithm, in place of simulated annealing, in the Box algorithm. The improved Box algorithm compares outputs derived by simulated annealing with outputs derived by the genetic algorithm and chooses the better set of values. We thus provide a tool to guide the modification of biological pathways that is both accurate and easy to apply. We illustrate our technique using the TNFα-mediated NF-κB pathway. INTRODUCTION Modeling and control of biological pathways is of extreme importance for studying mutations (Kitano, 2002) and for synthetic biology (Andrianantoandro et al, 2006). Parameter extraction (Feng and Rabitz, 2004) and development of mathematical models of biological systems have increased the possibilities for modeling pathways and for controlling bio-chemical reactions to yield desired protein output concentrations. Optimization and control of biological pathways involves changing model parameters such as rate constants and species concentrations to achieve desired levels of the output protein within a specified time interval. Typically, control is achieved by either random lab experiments (Guet et al, 2002) or directed evolution (Yokobayashi et al, 2002). We wish to supplement wet lab experimentation with faster and cheaper computer simulation, greatly reducing the required number of wet lab experiments and guiding researchers to choose those wet lab experiments most likely to yield desired results. Treating biological pathways as computational problems and applying mathematical modeling provides an efficient method to study how to modify them (Voit, 2000). Previous approaches to modeling were strictly based on mathematical concepts and did not include experimental constraints (Li et al, 2002; JDesigner, 2006; JSim, 2006). We have designed our Box algorithm to incorporate experimental constraints, producing