Optimization of Algorithms for Modeling Protein Structural Transitions from Sparse Long-Range Spin-Label Distance Constraints

Abstract

Abstract Function-related structural transitions of proteins are often large-scale conformational changes that are related to essential dynamics of a protein, which in turn proceeds along a small number of slow normal modes. Hence, it should be possible to characterize such transitions by an equally small number of distance constraints. These constraints should contain the information how the backbone coordinates move along the active normal modes. Such an approach based on residue-level elastic network models for the protein backbone is optimized with respect to the fit algorithm, constraint selection, and parametrization of the elastic network model. A stable fitting algorithm can be based on energy equipartitioning among the modes in the active space. This stabilization allows for extending active space dimension beyond the number of available constraints, which improves fit quality for transitions with a normal mode spectrum of only slowly increasing frequencies. In constraint selection, discrimination between the active normal modes appears to be more important than achieving large distance changes between initial and final structure. Parametrization of the network model has only a small influence on fit quality, as long as scaling of force constants with the inverse sixth power of the distance between network nodes is maintained. Elastic network models with a uniform force constant below a cutoff distance perform significantly worse. With 50 distance constraints, the optimized approach covers more than 50% of the structural change for 44% of all test cases, between 25 and 50% for 22% of the cases, and it fails for 33%.