GENETIC ALGORITHMS IN LOW RESOLUTION
PROTEIN STRUCTURE DETERMINATION
E. Pantos, P. Chacon, F. Moran, J.F. Diaz, J.M. Andreu
CLRC, Daresbury Laboratory
A genetic algorithm for iterative fitting of SAXS
data is described. A population of genes codifying a given mass
distribution is randomly generated. From each genotype information
the corresponding model structures can be processed by a SAXS
simulation procedure and compared to experimental data. The
reproductive fitness is used to generate the next population by
mutation and two-point crossover recombination using an elitist
strategy to enhances the GA performance. The algorithm described
produces fast convergence to a fittest model mass distribution. This
method affords a dramatic reduction of memory and processor time
required by other SAXS fitting methods. The effectiveness of the
procedure is demonstrated with synthetic objects, and by deriving a
16nm resolution model of a known protein structure from the
corresponding computed SAXS profiles.
(posted 15-Oct-97 jw)