Searching for chemical reaction mechanisms with genetic algorithms

This dissertation presents the application of genetic algorithms (GAs) to mechanistic problems in two areas of chemistry. GAs "evolve" better and better solutions to a given problem by carrying out an analogue of natural selection on a population of trial solutions, represented by "genetic" sequences of numbers.;The first application is the determination of good values for a set of parameters governing how enzymes in a simple biochemical model are regulated by end-products of the overall pathway. The "goodness" of a parameter set is given by an abstract functional goal to be carried out by the model, the proper direction of biochemical flux according to a biochemical "need". The formulation of this abstract goal into a concrete numerical objective function is given. A GA is then used to search for parameter values that give large values for this function, and hence for mechanisms good at carrying out the flux-direction task. The GA findings are largely consistent with intuitive predictions, namely that the system should have negative feedback and that the effect of the end-products be reciprocal.;The second application is the determination of a complete reaction mechanism for an oscillating chemical system starting with only a small amount of information. The genetic representation of complete reaction mechanisms and the proper formulation of selection criteria are discussed in detail. Criteria may range from a simple requirement of oscillation to matching specific period, amplitude, and phase properties of the system for which a mechanism is sought. An illustration is given using a simple model oscillator. When given information about the five species of the model, the GA procedure finds many oscillating mechanisms, but all of these are stoichiometrically inconsistent. A modified procedure incorporating a stoichiometric constraint finds consistent oscillating mechanisms after information about two of the four reactions of the model is given. All of the found mechanisms have the same structure as the model, and when more stringent selection criteria are used, the found rate coefficients match those of the model. This technique can replace the traditional method of the educated guess in proposing chemical reaction mechanisms.