Screening a large database of organic compounds for potential ligands to a protein is often seen as a simple extension of the docking problem, which is the prediction of the favorable binding mode for a single ligand. When doing ligand screening by docking, as in our screening tool Slide, the docking problem must be solved for each ligand candidate in the database. But, because hundreds ofthousands ofligands are screened, the time that a screening tool can spend for each compound must be far less than several minutes, which is the typical runtime for fast docking tools that model full ligand flexibility [1-3]. When spending only one minute per compound, the screening time for a database of 100 000 molecules is about 10 weeks. In order to

* To whom correspondence should be addressed. E-mail: [email protected].

reduce the runtime to a realistic time frame, say, a day, it is first important to efficiently rule out infeasible candidates, then to focus on the few promising molecules in the database.

The limitation of the time for conformational search in a screening tool also affects its scoring function, which is used to rate the complementarity of protein and ligand in a given conformation. Rather than estimating the binding affinity, a computational intensive and still imprecise science, the goal of the scoring function in a screening tool is to give an appropriate relative ranking for the potential ligands, with known or new 'real' ligands obtaining top ranks. Ideally, such a scoring function should be robust, with real ligands obtaining high scores irrespective of their exact binding mode. Another important and often neglected aspect of docking and database screening is the induced shape complementarity of the protein upon ligand binding. Many cases are known in which the binding site undergoes significant conform-ational changes when binding different ligands [4-6]. When assessing the quality of docking tools, typically known ligands are redocked into fixed binding sites that are tailored to bind that very ligand, since they are taken from the corresponding crystallographic complex. Although this is likely to bias the selectivity for the known ligand and its score relative to other candidates, the effect might be minor for lead optimization, where similar ligands are docked to compare their binding modes and relative affinities. However, when screening compounds from a database for lead discovery, bias towards known ligands should be avoided in the search. Our approach is to screen using the ligand-free conformation of the target protein, when available, and to model induced complementarity for the protein side chains as well as ligand when screening and docking a large variety of potential ligands.

In this article, we describe applications of our screening tool Slide [7], which is able to reduce large compound databases of more than 175 000 organic molecules to a ranked list of approximately 100 docked potential lig-ands within an hour to two days, depending on the binding-site characteristics and degree of flexibility in the screened molecules. In addition to ligand flexibility, SLIDE models full flexibility of ligands and protein side chains when docking potential ligands, and uses Consolv [8] to predict water-mediated interactions with the ligand.

0 0

Post a comment