Mathematics Colloquia and Seminars

Return to Colloquia & Seminar listing

Description

A large part of the increasing success of modern medicine over the past 50 years is due to the development of modern drugs. Drug development is amazingly expensive, however, and these costs as passed on to the consumer. One major bottleneck in the drug design process is that there is currently no reliable way to predict, given a target protein structure, which potential drug-like molecules (ligands) will bind with high affinity to a disease-associated active site. Thus the early stages of drug design typically amount to simple trial and error for a large library of molecules to identify potential drug molecules. Drug companies also employ so-called computational Docking methods to assist with this stage. In Docking, a large library of potential binders is screened computationally to try and identify those which are most likely to bind well for experimental testing. All of the major drug companies use these methods, but, while fast, they are unfortunately quite unreliable. Measured binding affinities actually have a stronger correlation with molecular weight than with the predictions of most major Docking packages. Docking makes so many approximations that it is difficult to know how to improve its results.

I will discuss my work to accurately calculate binding affinities with the highest reasonable level of theory for a simple test system, with a particular focus on cases where Docking predictions fail. My results show that current molecular mechanics force fields are capable of doing far better than Docking and not only accurately ranking potential binders but also accurately calculating relative free energies of binding and accurately identifying ligand bound orientations. Furthermore, I am working to learn from these detailed methods how to improve Docking. For example, Docking typically neglects most protein flexibility, and I am looking at how much this approximation would affect the results, to assess how many of Docking's failures are attributable to this approximation. Additionally, these results include contributions to binding affinity due to both protein and ligand entropies, which Docking currently neglects, and more sophisticated solvent models; I am working to systematically add back in some of the approximations Docking makes to assess the importance of these. I will also discuss plans for future work, including application of the method to systems of greater relevance to diseases, and ongoing experimental tests to verify some predictions I have made for some particular ligands.