Accommodating protein flexibility in computational drug design.

The need to account for the dynamic behavior of a receptor has long been recognized as a complicating factor in computational drug design. The use of a single, rigid protein structure—usually from a high-quality X-ray crystal structure— still is the standard in most applications (Zheng and Kyle, 1997; Walters et al., 1998). The choice to use a single protein structure is usually based on speed. For example, if a large database of compounds is to be screened for binding affinity, several conformers of each compound will be compared with each protein configuration. Although it is more accurate to use many representative protein configurations, it quickly becomes a very slow process to screen each conformer of each ligand against each protein configuration. It is usually impractical to attempt such prohibitively slow calculations because there is an unfortunate but necessary trade-off between speed and accuracy in computer modeling. Only recently have advanced methods been introduced to aid in a more accurate description of protein flexibility and its influence on ligand recognition. This has been aided by the exponential growth in the speed of computer processors, available RAM, and disk capacity—all of which are rapidly becoming less expensive. Here, we explain the need for accommodating an ensemble of protein configurations in drug design and the computational methods available for generating and manipulating that dynamic information. Most of the applications discussed below are improvements in liganddocking or in the generation of pharmacophore models for database searching. Protein Flexibility and Its Influence on Ligand Binding