Docking and scoring--theoretically easy, practically impossible?

Structure-based Drug Design (SBDD) is an essential part of the modern medicinal chemistry, and has led to the acceleration of many projects, and even to drugs on the market. Programs that perform docking and scoring of ligands to receptors are powerful tools in the drug designer's armoury that enhance the process of SBDD. They are even deployed on the desktop of many bench chemists. It is timely to review the state of the art, to understand how good our docking programs are, and what are the issues. In this review we would like to provide a guide around the reliable aspects of docking and scoring and the associated pitfalls aiming at an audience of medicinal chemists rather than modellers. For convenience, we will divide the review into two parts: docking and scoring. Docking concerns the preparation of the receptor and the ligand(s), the sampling of conformational space and stereochemistry (if appropriate). Scoring concerns the evaluation of all of the ligand-receptor poses generated by docking. The two processes are not truly independent, and this will be discussed here in detail. The preparation of the receptor and ligand(s) before docking requires great care. For the receptor, issues of protonation, tautomerisation and hydration are key, and we will discuss current approaches to these issues. Even more important is the degree of sampling: can the algorithms reproduce what is observed experimentally? If they can, are the scoring algorithms good enough to recognise this pose as the best? Do the scores correlate with observed binding affinity? How does local knowledge of the target (for example hinge-binding to a kinase) affect the accuracy of the predictions? We will review the key findings from several evaluation studies and present conclusions about when and how to interpret and trust the results of docking and scoring. Finally, we will present an outline of some of the latest developments in the area of scoring functions.