Research | John Terhorst, Ph.D.

My research projects use computers to explore questions in modern chemistry that experimental methods often cannot answer. Specifically, we apply chemical theory, statistics, and informatics to study the relationship between the structure of organic molecules and their activity in biological systems. This “quantitative structure-activity relationship,” or QSAR, allows us to design molecules with certain desired pharmacological properties, with the ultimate goal of developing therapeutic agents targeting infectious, inflammatory, and hyperproliferative diseases.

A Fragment-Based Search for HIV Inhibitors using Linear-Response Property Prediction

The foundation of any QSAR analysis begins with the fragment library: a database of molecular fragments and their associated geometric, physical, and chemical properties. This project involves design and population of a fragment library for discovery of inhibitors of HIV-1 reverse transcriptase, a key enzyme in the retroviral reproductive pathway. Research articles published in 2011 and 2012 indicate that heterocyclic cores connected by various linker groups can bind irreversibly to an allosteric site on HIV-RT, but novel structures such as these have not been widely studied in a systematic fashion. Our goal is to design a large set of heterocycles, linker groups, and derivatives, then to characterize them as individual fragments and as combinations thereof. Once the library is complete, we will use it to design drug candidates that should bind to our target based on complementary geometries, torsions, polarities, and hydrogen bonding; docking our candidates into the allosteric binding pocket of HIV-RT will then allow us to rank our results quantitatively.

Continuum Solvation Models and Force Field Development for Computer-Aided Drug Design

My graduate work involved development of a novel method for rapid free-energy calculations with continuum solvent models. In my dissertation, an implementation of the generalized Born / surface area (GB/SA) solvation model with free-energy perturbation (FEP), including an approximation used in calculating the total Born energy of the system, is presented. Our approximation is based on the assumption that a significant number of pairwise energy calculations may be omitted with little-to-no impact on the total change in energy of the system after a Monte Carlo move because the impact of a moving atom on the Born radius of a distant atom is small. Thus, we structured our implementation of GB/SA in such a way that the Born energy between an unmoving pair of atoms is only recalculated after a move if the Born radius of either atom has changed by more than a specified threshold since the last accepted move. Prior benchmarks demonstrated that existing GB/SA methodologies were insufficient for the purposes of calculating free energies of binding, and FEP simulations with GB/SA solvation were too computationally expensive to be used with any practicality. With our approximation, improved efficiency was achieved while affording minimal error: the influence of our approximation on accuracy of free energies of binding was negligible, with any error introduced by the approximation falling well below the statistical error of the Metropolis Monte Carlo algorithm, and speed-up of up to 62% was observed. The conclusion is that with our approximation, GB/SA is a viable solvent choice for FEP of large systems. Comparison between GB/SA and TIP4P in a substituent scan was quantitative to qualitative, with free energies of binding usually in agreement within 1 kcal/mol, producing the same substitution pattern on a drug candidate found to give high anti-retroviral activity as predicted by previous simulations with TIP4P explicit water.

E/Z Energetics for Molecular Modeling and Design

Thermochemical data obtained from G3B3 quantum mechanical calculations are presented for 18 prototypical organic molecules that exhibit E/Z conformational equilibria. The results are fundamentally important for molecular design including the evaluation of structures from protein–ligand docking. For the 18 E/Z pairs, relative energies, enthalpies, free energies, and dipole moments are reported; the E-Z free-energy differences at 298 K range from +8.2 kcal/mol for 1,3-dimethyl carbamate to −6.4 kcal/mol for acetone oxime. A combination of steric and electronic effects can rationalize the variations. Free energies of hydration were also estimated using the GB/SA continuum solvent model. These results indicate that differential hydration is unlikely to qualitatively change the preferred direction of the E/Z equilibria.

Exploring Dihedral Torsion Profiles with Implicit Solvent Models

Molecular mechanics are central to applications in computer-aided drug design. Of particular interest is the ensemble of conformations defined by one or more dihedral torsions within a given molecule. Often it is the case that these torsions give rise to unique molecular shapes, which can affect a potential drug’s ability to bind (or not bind) to its intended target. For instance, if the conformation that is required for binding is too far uphill in energy from the molecule’s native conformation, then binding is unlikely to occur; but if structural modifications can be made intelligently that can “lock” a molecule into its binding pose or otherwise coerce a molecule into a new low-energy conformation, binding can be enhanced. Thus, this project involves examining the dihedral torsion profiles of drug-like organic molecules in an aqueous environment to evaluate, among other things, the effects of solvation and substitution on molecular shapes. Our goal is to use chemical simulations to characterize the torsional energies of derivatives of benzene, pyridine, pyrimidine, pyran, furan, thiophene, and other heterocycles, with and without the effects of solvation by water; comparing the two sets can give insight into how the different classes of molecules conform under biological conditions and what energies are involved in their conformational equilibria.