Computational approaches to fragment based screening

Polarization is an often - neglected term in molecular modelling, and this is particularly the case in docking. However, the growing interest in fragment - based drug design, coupled with the small size of fragments that makes them amenable to quantum mechanical treatment, has created new opportunities for including polarization, anisotropic electrostatics and realistic repulsion potentials in docking. We have shown that polarization implemented as induced charges can offer in the region of a 10-15% improvement in native docking results, as judged by the percentage of poses within a rather tight threshold of 0.5 or 1.0 Å, where accurate prediction of binding interactions, are more likely. This is a significant improvement given the quality of current commercial docking programs (such as Glide use d here). This improvement is most apparent when the correct pose is known a priori, so that the extent of polarization is correctly modelled, and scoring is based on force - fields that do not scale the electrostatics. The introduction of specific active - sit e water molecules was shown to have a far greater effect than the polarization, probably because of the introduction of 3 additional full charges, rather than introduction of smaller charge perturbations. With active site waters , polarization is more likely to improve the docking when the water molecule is carefully orientated using quantum mechanical/molecular mechanics (QM/MM) methods. The placement of such water molecules is a matter of great current interest; we have shown that the water molecule can be placed with some degree of reliability simply by docking with the ligand present, provided that the water makes good hydrogen bonding interactions (these are the very conditions under which it is desirable to include the specific active-site water). Anisotropic electrostatics and exponential repulsion for rigid fragments was investigated using Orient and compared to QM/MM methods, all methods merited further research. The general hierarchy is that native docking using Glide (with polarization)> QM/MM (with MM polarization)> Orient-based methods. Thus, we expanded the Glide (with polarization) dataset to include more realistic crossdocking experiments on over 5000 structures. RMSD analysis resulted in many examples of clear improvement for including polarization.