New theoretical techniques are being developed and characterized. These efforts are usually coupled with software development, and involve the systematic testing and evaluation of new ideas. By performing an autopsy of our results from the SAMPL4 prediction, we have performed a bench marking study using hydration free energy of 20 small molecules to establish an accurate protocol for combining density function theory calculations with the non-Boltzmann Bennett free energy method. This optimized protocol enjoyed an RMSD error of 0.89 kcal/mol with respect to experimental measurements, the best results published for the blind SAMPL4 data set. Furthermore, this protocol formed the basis of our best performing submission for the SAMPL5 challenge, which was the second most accurate submission in the entire challenge. We participated in the recent SAMPL5 blind prediction challenge, and calculated partition coefficients for 53 small drug like molecules. One of our submissions, based on a protocol optimized from data from a previous challenge, performed very well, ranking second overall by RMSD. In this work we also attempted to model the effects of protonation state and tautomerism, and include them in our blind predictions. The results of applying these protonation corrections was mixed. In all cases, including the effects of protonation increased the correlation of the ranking of the relative hydrophilicity of our results relative to experiment, however in many cases, our RMSD values increased as well. This mixed result indicates that while our corrections are capturing some physical effect, more work is required to refine this approach. Free energy methods with QM Theoretical approaches can be a useful partner for predicting physiochemical properties important in biomolecular recognition events as such predictive approaches can provide guidance in high-throughput screening for rational design. Although QSAR predictive models are often used for predicting various properties, the quality of these models are limited to what they are parameterized to fit and experience difficulty for determining properties coupled to changes in the electronic environment. The partition coefficients of a diverse set of 53 drug-like molecules of the SAMPL5 challenge were predicted using electronic structure methods. Several DFT functionals partnered with correlation consistent basis sets were evaluated for the prediction of the partition coefficients with the intent to identify which approach provides a balance between computing costs and accuracy. We demonstrate that DFT functionals with less parameters performs best and identify some of the biases associated with the implicit solvation model and the DFT functionals. Theoretical approaches can be a useful partner for predicting physiochemical properties important in biomolecular recognition events as such predictive approaches can provide guidance in high-throughput screening for rational design. Although QSAR predictive models are often used for predicting various properties, the quality of these models are limited to what they are parameterized to fit and experience difficulty for determining properties coupled to changes in the electronic environment. The partition coefficients of a diverse set of 53 drug-like molecules of the SAMPL5 challenge were predicted using electronic structure methods. Several DFT functionals partnered with correlation consistent basis sets were evaluated for the prediction of the partition coefficients with the intent to identify which approach provides a balance between computing costs and accuracy. We demonstrate that DFT functionals with less parameters performs best and identify some of the biases associated with the implicit solvation model and the DFT functionals. Binding free energy calculations and methods Improving the methods used for the characterization of solubility and binding affinity is a crucial step in advancing drug design. Free energy simulations (FES) have been used extensively in predicting binding affinities for receptor-ligand complexes. As part of SAMPL5 blind challenge, we used thermodynamic integration and Hamiltonian replica exchange with the Bennett acceptance ratio to evaluate the absolute binding free energy values for small molecular systems consisting of octa acids and ligands. Our estimations with low errors with respect to experimental data were consistently ranked near/at the top. This project was performed in collaboration with Prof. C. Seok at Seoul National University and Prof. A. Mackerrel at University of Maryland, Baltimore. Improved polarization methods Having developed a hierarchy of methods to treat electron polarization using perturbation theory, we turned our attention to a more generalized extrapolation scheme. After formulating the gradients and virials for any order of perturbation theory, we devised a scheme to find optimized coefficients for each order, greatly improving the approximation of infinite order expressions using only a handful of terms. All methods have been made available to the community in the CHARMM, TINKER and OpenMM program packages. Developing a new reaction pathway sampling method via global optimization of action We developed a new computational approach to sample multiple reaction pathways with fixed initial and final states through global optimization of Onsager-Machlup action using the conformational space annealing method. This approach successfully samples not only the most dominant pathway but also other possible ones without initial guesses on reaction pathways. The sampling ability of the approach was assessed by finding pathways for the conformational changes of alanine dipeptide and hexane. The benchmark results on alanine dipeptide identified 8 possible pathways for the conformational change between C7 axial and C7 equatorial states of alanine dipeptide. It was also demonstrated that the rank order of actions and transition time distribution of multiple pathways identified by the new approach are in good agreement with those of molecular dynamics simulations. Conformational transition pathways between the all gauche(-) and the all gauche(+) states of hexane were sampled using the new method. The results show that the lowest action pathways, the most dominant pathways, are consistently found in multiple independent simulations without initial guesses on pathways. Currently, we are testing the new method on finding folding pathways of small proteins. MESS-QM/MM: Fast Calculations of QM/MM-Corrected Hydration Free Energies for Rigid Solute Molecules We developed and applied the Multiple-Environment Single-System (MESS) QM/MM methods to the hydration free energy calculations for SAMPL4 molecules. The QM/MM portion of the computational time was reduced 50-200 times, leading to essentially the same results without the MESS-QM/MM approximation. Ab initio QM/MM with Particle Mesh Ewald We are implementing an improvement for electrostatic treatment of complex systems with a rigorous QM/MM/PME method that does not involve calculating partial charges nor Mulliken populations. We directly place and use QM charge density on the PME grid instead of first calculating intermediate partial charges to represent QM electronic density. This PME extension will provide more accurate results because of its consistency with PME grids and by the removal of an unnecessary approximation. This new approach will also be fully consistent with our use of classical multipoles and induced polarization.
Showing the most recent 10 out of 63 publications
