Modeling DNA structures

For a molecule of biological importance, one expects a strong correlation between the three-dimensional structure and its biological function(s). Molecular simulations allow the prediction of physical properties of macromolecules. Many of these properties are closely related to the molecular structure. Model-building studies may thus supplement structurally low-resolution experimental data with detailed three-dimensional hypothetical atomic models. Such studies may give a consistent integral view of a wealth of experimental data. In most cases, such models will predict the outcome of certain experiments. Their actual results will often either confirm the model, be used for further refinement, or demand a major revision. Empirical force-fields provide a large amount of physicochemical knowledge concerning structural and other physical properties about various classes of molecules. They give reasonable bond distances and angles and prevent short van der Waals contacts. Difficulties arise for the prediction of large-scale structural elements. These are not only determined by short-range interactions, but also result from long-range electrostatic, hydration, and hydrophobic forces. In the case of the ionically driven DNA B-Z transition, the point of transition as a function of ionic strength and size can be correctly predicted (33). More must be done for a better understanding of the hydration forces (6). What type of questions will be reliably answered by a force-field? Relatively safe answers concern the local geometry of the molecules. If a conformation leads to strong distortions of bond distances or angles or to close van der Waals contacts, it can safely be rejected. Optimizing such unfavorable structures energetically may lead to structures showing how to avoid such distortions. More difficult are energetic questions: Which of two conformers is more stable, or what is the free energy of the substrate in the active site (63)? One cannot always be sure that the force-field provides the correct answers. Therefore, one should concentrate on questions that can be checked experimentally. The application of such concepts to model curved DNA (19, 60) and the DNA four-way junction (61) provides promising results. To explore the knowledge contained in the force-fields, several methods have been proposed. Taking advantage of structural symmetries may improve critically the convergence, while refining the target molecule or its building blocks. A numerically stable derivative for the torsion potential has been proposed. The optimization method of conjugated gradients (see Section II,B) is a powerful tool to find the way downhill toward a local minimum. To surmount barriers and escape local minima requires nonlocal optimization procedures.(ABSTRACT TRUNCATED AT 400 WORDS)