Optimization of a genetic algorithm for searching molecular conformer space

Abstract

We present two sets of tunings that are broadly applicable to conformer searches of isolated molecules using a genetic algorithm (GA). In order to find the most efficient tunings for the GA, a second GA-- a meta-genetic algorithm--was used to tune the first genetic algorithm to reliably find the already known a priori correct answer with minimum computational resources. It is shown that these tunings are appropriate for a variety of molecules with different characteristics, and most importantly that the tunings are independent of the underlying model chemistry but that the tunings for rigid and relaxed surfaces differ slightly. It is shown that for the problem of molecular conformational search, the most efficient GA actually reduces to an evolutionary algorithm.