Background

The majority of results reported for database screening are based on the application of tools that were designed to predict the favorable binding mode and the binding affinity for a single ligand. Several docking tools are available, and they are widely used for structure-based ligand design [2,3,9-19].

Docking tools can be classified by the method they use to represent the binding site, by the technique for sampling ligand conformations, and by the way they construct the docked ligand. All docking tools fast enough to screen a large data set of molecules are based on so-called descriptor-matching approaches [20], which means that they represent the binding site by a template of points, onto which ligand atoms are mapped during the search. The template points can describe the shape of the binding site [9,21], or favorable chemical interaction centers above the protein surface, where hydrogen-bond donors or acceptors, metal ions, or hydrophobic groups of the ligands can be placed [1-3,11,22]. During the search for the optimal binding mode of a ligand, different conformers for the molecule are generated, which can be done randomly, e.g., by using a genetic algorithm [16,17,19,23], or by systematically sampling discrete torsional angles for the rotatable bonds of the ligand [1-3]. The faster docking tools construct the ligand incrementally in the binding site [2,3], or dock fragments of the ligand independently and chemically link them later, if the combination is feasible [1,12-14,24]. All docking tools that have been used for database screening employ a bindingsite template for guiding the search and incremental construction ofthe ligand in the binding site [1,25-30].

While all recent docking tools consider full ligand flexibility, induced complementarity of the protein upon ligand binding is not modeled, at least not in the faster docking tools. In approaches that model protein flexibility, this is often limited, e.g., by only rotating terminal hydrogens to optimize intermolecular hydrogen bonding [16,17], or by using rotamer libraries for the side-chain conformations [31,321. Other approaches model explicitly defined side-chain flexibility [33], hinge bending [34], or they dock ligands against an ensemble ofprotein structures [35]. Molecular dynamics simulations [3638] may yield the most realistic models of protein and ligand flexibility, but the resulting runtime for docking a single ligand is likely to be in the range of hours.

A drawback of many docking tools is that they neglect the effect ofbinding-site solvation and the potential for water-mediated interactions between protein and ligand [8,39-42]. While it is certainly possible to consider bound water molecules as part of the rigid protein in most docking tools, recently three sophisticated approaches have been reported, which either predict conserved binding-site waters [8], compute potential water positions prior to docking [43], or solvate the ligand molecule [29].

Recent docking and screening tools can identify potential ligands from up to 150 000 compounds within a few days, when considering full ligand flexibility [1,25-28,44]. Even in the absence ofmodeling inducible complementarity and solvation, there have been successful project reports in structure-based lead discovery or design, including the identification of new inhibitors for thymidylate synthase [25], P. carinii dihydrofolate reductase (DHFR) [27], P. falciparum DHFR [45], trypanothione reductase [46], and human thrombin [47]. Our goal with the new screening tool SLIDE is to incorporate a balanced model ofprotein and ligand flexibility as well as a knowledge-based model of solvation that is fast enough to be used for screening and docking hundreds of thousands of compounds.

0 0

Post a comment