In silico screening methods for chemical libraries

Faced with a locked door to which you don't have a key, you have two options: either patiently construct a key step by step, or search for one in a bag of existing keys to match the lock on the door to be opened. This latter "emergency locksmith" strategy sums up the fundamental principle of in silico (virtual) screening applied to the design of active molecules. Rather than creating an ideal molecule that has yet to be synthesized, it may be more interesting simply to select the existing molecule(s) that meet the required specifications. Initially dedicated to the prediction of protein-ligand interactions (which molecules are likely to bind to a protein of interest?), computer screening of chemical libraries has considerably broadened its scope of application to the prediction of any property of interest, not only biological or pharmacological, but also physico-chemical. The considerable development of real or virtual chemical libraries gives this approach unrivalled value for money when it comes to prioritizing a small number of molecules whose properties are intended to be validated experimentally. In fact, it is pointless and costly to evaluate millions of molecules experimentally if the IT tool is capable of accurately selecting the few hundred or so molecules worth testing. Of course, selection is never perfect, and we'll be analyzing the reasons why. Nevertheless, virtual screening of millions of molecules remains accessible to academic and/or private structures lacking the financial resources required for experimental screening, as systematically used by the pharmaceutical industry. Its great merit therefore remains the exploration of a huge chemical diversity at low cost. Looking back over the past ten years, the various methods have also matured, and their field of application is constantly expanding.

All virtual screening is based on the choice of a chemical library and a screening method suited to the problem at hand. As we shall see, many combinations are possible, and the initial choice has a direct influence on the quality of the results obtained. Secondly, we present a number of case studies illustrating not only the potential of the various methods, but also their limitations.