Rapid estimation of relative binding affinities of enzyme inhibitors

Abstract

Computer-assisted molecular modelling (CAMM) comprises a variety of computational methodologies intended to quantitatively or qualitatively describe molecular properties. In some cases, the field has advanced to a state where accurate predictions are possible (e.g. geometric and electronic properties of small molecules). In other cases, however, the properties are complex and require either advances in theory or substantial increases in computational power (e.g. protein folding). One application of CAMM that has received considerable attention over the past two decades entails its use as an aid in drug-design. Ideally, CAMM would provide rapid and accurate prediction of drugtarget binding affinities such that a large and structurally diverse population of potential targets could be evaluated and thereby prioritized prior to chemical synthesis. In reality, methodologies have been advanced that either provide qualitative rank ordering of a large number of molecules in a relatively short period of time or, at the other extreme, generate quantitatively accurate predictions of relative binding affinities for structurally related molecules using substantial computing power. Consequently, techniques that increase speed without greatly compromising accuracy (or vice versa) are of value to drug-discovery programs. Advances in protein crystallography and molecular simulations have greatly aided computer-assisted drug-design paradigms and the accuracy of their binding affinity predictions [1,2]. Methods of inhibitor design range from graphical visualization of the ligand in the binding site to calculation of relative binding affinities using molecular dynamics simulations in conjunction with the TCP approach [3,4]. Figure 1 shows a flowchart employed by drug-discovery groups for structure-based drug-design. Typically, the process begins by generating a working computational model from crystallographic data which includes the development of molecular mechanics parameters for non-standard residues, building any missing segments, assigning the protonation states of histidines, and orientation of carbonyl and amide groups of Asn and Gln amino acid residues based upon neighboring donor/acceptor groups. Inhibitor design is then aided by a variety of visualization tools. For example, hydrophobic and hydrophilic regions of the active site are readily identified by calculating the electrostatic potential at different surface grid points. Frequently, the information gained on the characterization of the active site is supplemented by analyses of ligand conformational energies. These studies are often followed by an estimation of binding affinity which is performed at three different levels of complexity (Fig. 1) depending on computational power, time and resources, namely: (i) qualitative predictions based on docking/ visualization and molecular mechanics calculations; (ii) quantitative predictions based on