Abstract
If the structures of proteins are modeled by hard balls representing each alpha-carbon atom then the maximum overlapped volume of those balls provides a measure of the degree of similarity between a pair of proteins. This paper describes the use of a genetic algorithm (GA) to calculate such similarity measures, with experiments being carried out that matched individual protein structures both against themselves and against other proteins. These experiments demonstrated that the best results were obtained by parameterising the GA as follows: using an exponential distance measure with a coefficient value of -0.1 as the fitness function in a steady-state GA with 100 chromosomes, a selection pressure of 1.5 and a crossover probability of 40%. The resulting similarities are highly correlated with those resulting from the allover method for calculating inter-protein similarities.
