Ron Wehrens & Lutgarde Buydens
Since the first papers of Genetic Algorithms (GA’s) for the optimisation of molecular structure, now more than 10 years ago, an amazing amount of applications in this field have been published. Several reasons can be identified. First of all, GA’s are easily applicable to optimisation problems where a measure of the quality of a trial solution is available. Using freely available toolboxes, only a minimum amount of programming is required to tackle dicult problems. Secondly, standard settings in many cases work well. Although many parameters can be tuned in genetic algorithms, this is often not necessary. Thirdly, and perhaps most importantly, the concept of “the survival of the fittest” that lies behind the Genetic Algorithm is appealing, and easily explained to non-experts.
In this paper, an overview of applications will be given, where the stress is on the geometry optimisation of small molecules. Applications can be categorised according to the quantity that is optimised: either the internal energy as calculated by some force field, or the agreement with experimental data such as spectra. In the latter case one either tries to fit the spectra directly, or to fit physico-chemical parameters derived from the spectra. Standard GA’s can be improved upon in several ways. Examples are methods to prevent premature convergence such as sharing or crowding, or to evolve several populations in parallel. A particular characteristic of optimising molecular structure is that often it is not only the one best conformation that we are interested in, but rather a whole group of structures that exist at room temperature. Criteria that take this into account have been used to fine-tune GA settings for the optimisation of small organic molecules.