Anautomatedprocedurefor protein designbyoptimization of a sequence-structurequalityhasbeendeveloped. Themethodselectsa statistically optimalsequencefor a particularstructure,ontheassumptionthat sucha protein will adoptthe desiredstructure. Wepresenttwooptimization algorithms:oneprovidesan exact optimization whilethe otherusesa combinatorialtechniquefor comparativelyrapidresults. Bothare suitablefor massively parallel computers.Aprototype systemwasused to design sequenceswhichshouldadopt the four-helix bundleconformationof myohemerythrin.Theseappear satisfactoryto secondarystructureandprofileanalysis. Detailedinspectionrevealsthat the sequencesare generally plausiblebut, as expected,lacksomespecificstructural features. Thedesign parametersprovidesome insightintothe generaldeterminantsof proteinstructure.
Steven E. Brenner, Alan Berry