Conditional optimization: a new formalism for protein structure refinement.

Conditional optimization allows unlabelled loose-atom refinement to be combined with extensive application of geometrical restraints. It offers an N-particle solution for the assignment of topology to loose atoms, with weighted gradients applied to all possibilities. For a simplified test structure consisting of a polyalanine four-helical bundle, this method shows a large radius of convergence using calculated diffraction data to at least 3.5 A resolution. It is shown that with a new multiple-model protocol to estimate sigma(A) values, this structure can be successfully optimized against 2.0 A resolution diffraction data starting from a random atom distribution. Conditional optimization has potentials for map improvement and automated model building at low or medium resolution limits. Future experiments will have to be performed to explore the possibilities of this method for ab initio phasing of real protein diffraction data.