Fragment Based Pose Prediction and Affinity Scoring
Monday, May 4th, 2009Fragment based drug design (FBDD) approaches to inhibitor development often include in silico surveys of the binding preferences of small ligands against the target binding. There are instances of cases where FBDD has been applied in the course of drug development as well as retrospectively with truly useful outcomes.[1] High throughput docking approaches that exhaustively probe the binding site can be useful if they are validated as to their ability to reproduce known structural biology solutions delineating preferred fragment poses and binding affinities. A number of careful benchmark problems exist for calibrating docking and scoring approaches to this task. One of them is the series of papers by Stoichet and coworkers who `developed a small’ well characterized cavity in T4 Phage Lysozyme via mutations at sequence positions 99 and 102. In particular the, L99A and M102Q mutations were used to create a cavity, fragment libraries were then screened against the target binding site, crystallographic characterizations of bound fragments made, and predicted and actual binding affinities compared.[2,3,4] Figure 1A shows a depiction of the binding site with 2-allyl-6-methyl-phenol bound. Hydrophobic contacts complimented by 1 hydrogen bond determine the x-ray pose. While Figure 1B shows some of the characteristic fragments employed in one of the studies
Figure 2A shows the eHiTS pose prediction pose accuracy for a small fragment series (shown in figure 1B) with the closest and top rank poses depicted. Note that four of the six cases gave top-rank docking poses of ~0.5 angstrom. While two of the six cases are not acceptable with top-rank values ~2.5 angstroms, simple family training on the series leads to the pose accuracy profile comparison shown in figure 2B. The results illustrate two facets: 1) generally speaking pose accuracy for fragment prediction for eHiTS will be sub-1-angstrom and 2) and simple family training (tuning) on 4-5 ligand/fragment bound crystal structures improves the pose prediction. This weight tuning procedure while often unnecessary, is accomplished in a matter of minutes with the eHiTS Tuning package that enables the user to develop a customized scoring function.
One also wants to be able to predict affinities of fragment libraries to a target binding site. This is a challenging problem and generally speaking docking protocols perform better at pose prediction than affinity scoring. Nevertheless, at the same time one is attempting to screen fragments for their probable binding preferences in the cavity it would be useful to have at least a semiquantitative ranking. A classic example of fragment based screening via x-ray crystallography to discovery of inhibitors lies in the work of Congreve and coworkers developing inhibitors of beta-secretase.[5] A series of fragments and refinements on the path of optimization are shown in figure 3 along with their binding affinities/molecular weights and logP values.
Figure 4 Panel A) shows the IC50/score correlation for eHiTS and Panel B) a molecular mechanics Poisson-Boltzmann scoring of several of the fragments/ligands. The eHiTS Score ln(IC50) correlation had a correlation of R2=0.61 and a Pearson coefficient of 0.78 while the MMPBSA scoring had an R2=0.51 and a Pearson coefficient of 0.72. The eHiTS scoring took on the order of 7-minutes for this set while the MMPBSA including charge/parameterization/simulation took several hours. While `affinity estimation’ via docking scoring functions can only be approximate, the eHiTS scoring function is adequate to the task. Figure 4B highlights the fact that the eHiTS Scoring has a good correlation with the enthalpic portion of the MM-PBSA free energy. This short synopsis of more detailed in-house studies illustrate the manner in which eHiTS docking and scoring is a good underpinning to a fragment de novo based approach to inhibitor design.
REFERENCES
[1] Congreve M, Chessari G, Tisi D, Woodhead AJ., “Recent developments in fragment-based drug discovery.”, J.Med.Chem., 51:3661-80 (2008).
[2] Wei, B.Q., Baase, W.A., Weaver. L.H., Matthews, B.W., Stoichet, B.K. “A Model Binding Site for Testing Scoring Functions in Molecular Docking.”, J. Mol. Bio. 322:339-355 (2002)
[3] Wei, B.Q., Baase, W.A., Weaver. L.H., Ferrari, A.M., Matthews, B.W., Stoichet, B.K., “Testing a Flexibile-receptor Docking Algorithm in a Model Binding Site.”, J.Mol. Bio. 327:1161-1182 (2004).
[4] Graves, A.P., Shivakumar, D.M., Boyce, S.E., Jacobson, M.P., Case, D.A. and Stoichet, B.K., “Rescoring Docking Hit Lead Lists for Model Cavity Sites: Predictions and Experimental Testing.” J.Mol.Bio. 377:914-934 (2008).
[5] Congreve M, Aharony D, Albert J, Callaghan O, Campbell J, Carr RA, Chessari G, Cowan S, Edwards PD, Frederickson M, McMenamin R, Murray CW, Patel S, Wallis N.,” Application of fragment screening by X-ray crystallography to the discovery of aminopyridines as inhibitors of beta-secretase.” J Med Chem. 50:1124-32 (2007).
Posted by Dan Harris
