One major emphasis is binding prediction
A major focus is the binding of small-molecule ligands to proteins. While current computational methods see widespread use in the pharmaceutical industry in drug discovery applications, accuracy is limited and these approaches fall far short of the goal of using computers to suggest new drug candidates. Methods we recently developed and applied have achieved far greater accuracies at computing and even predicting binding affinities than previous methods, so we are working to begin applying these in more complicated and pharmaceutically relevant binding sites. Projects involve both applications to drug discovery problems, and methodological improvements. Our work in this area focuses on using so-called alchemical free energy techniques for predicting binding affinities using molecular simulations. (See alchemistry.org for more.)
We continue to advance methods for binding prediction using model binding sites:
Previously, we pioneered techniques for computing absolute binding free energies, applying these first in several model binding sites — an apolar binding site in T4 lysozyme, and then in more polar version of the binding site which introduces hydrogen bonding. This included work on handling uncertainty in ligand binding modes. Follow-up work dealt with how protein conformational changes adversely affect relative free energy calculations. Continuing the progression towards biological binding sites we added charge into the mix, studying two different charged designed binding sites in Cytochrome C peroxidase, which involve some new methodological challenges. We continue to work in the space of model binding sites, and are very interested in finding suitable model systems to bridge between these sites and systems of current interest in pharmaceutical drug discovery, which are often not particularly suitable for benchmarking new methods. We also hope to take the SAMPL series of blind challenges in this direction.
We apply binding free energy techniques
We have several more application-oriented problems of binding free energy techniques. For example, we are collaborating with the Poulos group (UCI) studying nitric oxide synthase inhibitors as antibacterial targets and, separately, for neurodegenerative diseases. We also have collaborations with pharmaceutical partners exploring ligand binding mode sampling in a fragment based drug discovery setting, and we are collaborating with the Tobias group (UCI) in studies of several ion channels. In one fo these, we are studying channel blockers, also in collaboration with Francesco Tombola (UCI).
Hydration free energy calculations and solubilities
Hydration free energies are important for several reasons. One is simply that it is now possible to calculate these extremely precisely for small molecules, enabling quantitative comparison between simulations and experiment. They are also thought to provide some idea of the level of accuracy that can be expected in drug discovery applications. Finally, they provide a probe of the underlying physics of hydration. These reasons make them rather interesting for computational studies.
Previous work on hydration free energies
We have a long history looking at hydration free energy calculations using alchemical free energy techniques, beginning with testing a variety of different charge models and two water models for hydration free energies of around 40 different molecules (see also follow-up). Later, we did blind predictions of hydration free energies on several different occasions, most recently just a few months ago. We also looked at solvation of more than 500 different molecules in both explicit and implicit solvent, providing what is becoming a standard dataset for hydration free energies whih we have expanded into the FreeSolv database (which we just updated and revised). We have also highlighted how water asymmetrically solvates solutes of opposite polarity. However, new experimental hydration free energies are becoming a rarity, so we are moving in the direction of other, related properties as tests and benchmarks for forcefields (and fodder for the SAMPL series of blind challenges), including relative solubility, partition coefficients, and distribution coefficients (see also the SAMPL6 distribution coefficient challenge we coordinated).
Ongoing work in this area includes using hydration free energies to help develop better force fields (for example, we are finishing work with collaborators on a new set of hydroxyl parameters) and test existing force fields and as a test bed for enhanced sampling techniques. We are also working on applying hydration and solvation free energy calculations in design problems. Further, these calculations are part of a solubility calculation, so we are also collaborating on techniques for physical calculation of solubilities.
The Open Force Field Initiative
We’re very excited to finally see a path forward towards better force fields for molecular simulations, and a better way towards developing them. This is a collaboration with Michael Shirts (Colorado), John Chodera (MSKCC), and Christopher Bayly (of OpenEye Software). Bayly spent Summer 2016 in the Mobley Lab helping us get off the ground to use SMIRKS as a language for chemical perception for force fields, and we now have a new SMIRNOFF (formerly SMIRFF) format for force fields which assigns parameters directly based on molecular chemistry, rather than indirectly via the intermediate step of atom typing. This fixes a number of problems with earlier force fields, and opens up a wide range of possibilities we will discuss in more detail soon. This also fits into the bigger picture of our effort by giving us a language we can use to sample over force field chemical perception in the context of parameter fitting. The end goal of all of this is to develop an automated Bayesian machinery for force field development, where force fields are no longer developed by human experts over a period of many years, but instead, there is debate about what ingredients go into a force field (functional form, data to fit to, etc.) but once this is determined, force fields are constructed via a fully automated process. Additionally, the Bayesian formalism we are working towards will allow us to produce parameter uncertainties, providing model uncertainties in force field predictions that have previously been unquantifiable. While this may be a long road, we’re excited about the progress already made and believe this will enable force field science in a way which has never been possible previously.
Methodology, Sampling Techniques, and Tools
Previously, we highlighted the importance of soft core techniques for binding free energy calculations, and the importance of long-range Lennard-Jones interactions in obtaining free energy estimates that are robust with respect to the choice of LJ cutoff. We have also developed a framework for including free energies of slow protein conformational changes in binding free energy calculations. More recently, we collaborated on enhanced sampling techniques for improved conformational sampling of small molecules using expanded ensemble techniques. Ongoing work focuses on improved methods for computing the relative free energies of ligand orientations in binding sites via nonequilibrium candidate Monte Carlo (NCMC) via our BLUES package, which is under development and includes methods for better sidechain sampling in torsions and rotational degrees of freedom; and general strategies for better sampling of slow degrees of freedom in free energy calculations.
We make free energy calculations easier to apply:
A major bottleneck in using these calculations in a drug discovery setting has been the difficulty of setting them up, and the expertise required to do so. We are working on new tools to make free energy calculations dramatically easier to apply, including a tool for automated planning of relative free energy calculations, Lead Optimization Mapper (LOMAP, GitHub). .