Our paper describes a methodology for carrying out high throughput docking of ligands into protein receptors. In the paper, advantages are demonstrated as compared to alternative approaches in the literature. There is widespread interest in the pharmaceutical and biotechnology industries, as well as in academia, in better approaches to virtual screening of receptors. The citations reflect this interest as well as the fact that a significant number of groups are currently using the computer program, "Glide," in which the methodology is embodied.

As indicated above, the paper describes new algorithms and scoring functions for virtual screening, and the computer program in which this methodology is implemented is available for distribution. Improved prediction of the structure of protein-ligand complexes, as compared to other widely available programs, is demonstrated in the paper, and the ability to discriminate active compounds from a comparison database of random, drug-like molecules, is demonstrated for a wide variety of pharmaceutically relevant receptors. This methodology is useful to those interested in obtaining novel lead compounds for drug design via virtual screening. It is also useful in a lead optimization context in predicting the structures of protein-ligand complexes for which experimental data is not available.

I have a long-standing interest in the use of physical chemistry-based methods for drug discovery. Several years ago, after reading a variety of papers in the literature, I thought of some ideas for improving the speed and accuracy of the calculations, and, in view of the central role that high-throughput docking was beginning to play in the pharmaceutical industry, decided to try to implement and test these ideas. The work began in my laboratory at Columbia and then developed into a collaboration with the company Schrodinger, Inc., which distributes a commercial version of Glide. Over time, the project has evolved into a sophisticated methodology for handling large databases of compounds in a robust fashion, in addition to continuing the development of novel algorithms and scoring functions aimed at improving the effectiveness of virtual screening—a technology that is still in its infancy.

Computational methods promise to dramatically reduce the cost of drug discovery by enabling the interaction of candidate drug molecules with their protein receptors to be modeled on the computer. The key barrier to achieving such cost reductions is the accuracy of the computer models. Our paper describes significant improvements in accuracy as compared to previous methods in the literature. While there is still much work to be done, rapid progress by our group and others suggests that computationally-driven drug design will assume increasing importance over the next decade.