The potentials of conditional optimization in phasing and model building of protein crystal structures

Abstract

Model building is a pivotal step in protein-structure determination, because with an atomic model available the vast amount of geometrical prior knowledge may be applied to the structure-determination process. Here, conditional optimization, a method that does not require interpretation of the electron-density map, is described. Instead, this method refines loose atoms for which all chemical interpretations are considered simultaneously using an N-particle formalism. This method bears the potential of introducing the geometrical data much earlier in the structure-determination process, i.e. well before an interpretable electron-density map has been obtained. Here, results from two tests are presented: automated model building of three proteins with diffraction data extending to 2.4-3.0 A resolution and ab initio phasing of a small four-helical bundle with diffraction data to 2.0 A resolution. Models built automatically by the widely used programs ARP/wARP and RESOLVE and those from conditional optimization per se, without discrete modelling steps, had comparable phase quality and completeness, except in loop regions, which are poorly modelled by the current force field in conditional optimization. Optimization of multiple random starting models by conditional optimization yielded models revealing the four helices of the four-helical bundle.