Conclusions

We have demonstrated that 'virtual NMR screening' was able to reproduce the experimental results described for an NMR-based screening method. Both structures and binding free energy were in agreement with experiment. The root-mean-square errors derived from correlation plots of the experimental vs. predicted binding affinities for FKBP and stromelysin were 0.5 kcal/mol and 1.5 kcal/mol, respectively. This demonstrates the ability of the HTS program to rank ligands that differ in affinity by approximately 10-fold or more.

As mentioned previously, protein flexibility could be a potential problem in any docking simulation that does not account for it. We have accounted for some of this flexibility by using a soft-core potential in both the docking and scoring energy functions to represent the local movements in the protein atoms, such as thermal fluctuations. Large scale motions, such as binding site formation upon complexation, are not included in the present study due to their computational expense. However, researchers in the docking community are actively working to develop computationally inexpensive methods to model large scale conformational motions which could be applied in this method.

Using 'virtual NMR screening' as computational aspect prior to experimental work could facilitate timely drug discovery. Our docking and scoring methods can dock and score a ligand with a few rotatable bonds within a few minutes or less. The HTS scoring method can differentiate between compounds with high and low affinities, yet due to its empirical nature, cannot distinguish between compounds that differ by approximately 1-2 kcal/mol. Such distinctions are better suited to other computational methods such as free energy perturbation or experimental methods such as focussed combinatorial library screening. 'Virtual NMR screening' is most appropriately used as a pre-screening tool to suggest possible leads for experimental testing.

0 0

Post a comment