Talk:2010 Workshop on Free Energy Methods in Drug Design

Talk schedule planning

(Apologies for the disorganization of this page -- it is still in intermediate planning stages.)

No titles yet

• Christopher Bayly, Merck-Frosst
• Christopher Chipot, Nancy Université/UIUC
• Richard Dixon, Vertex
• Jonathan Essex, University of Southampton
• Viktor Hornak, Merck
• Mark Murcko, Vertex
• Andrew Pohorille, NASA AMES
• Julia Rice, IBM Alamden
• Benoît Roux, U Chicago
• Thomas Simonson, Ecole Polytechnique, France
• William Swope, IBM Almaden

25 minutes

• Christopher Bayly (Merck Frosst)
• Scott Brown (Abbott) - What does it take to positively impact industrial drug design?
• Matt Clark (Pharmatrope) - Fast free energy calculation based on Monte Carlo sampling of pose and conformation with a flexible protein model - needs to speak on mon or tue
• Hideaki Fujitani (University of Tokyo) - MP-CAFEE: principles and challenges
• Emilio Gallicchio (Rutgers University) – The Binding Energy Distribution Analysis Method (BEDAM): Theory and Applications
• Marti Head (GSK) - Free Energy Simulations: and miles to go before we sleep
• William Jorgensen (Yale) - Successes and Challenges for FEP-Guided Optimization of Enzyme Inhibitors - needs to speak early in conference
• Julien Michel (Yale) - Water in binding free energy calculations
• Mark Murcko (Vertex)
• Andrew Pohorille (NASA AMES)
• Julia Rice (IBM Almaden Research)
• Régis Pomès (Hospital for Sick Children/University of Toronto) - Efficient Algorithms for Conformational Sampling in Free Energy Simulations
• Enrico Purisima (NRC Canada) - The Usual Suspects
• Pengyu Ren (UT Austin) - Free Energy Calculations using Polarizable Multipole-based AMOEBA Potential
• Sereina Riniker (ETH Zurich) - Efficient computation of relative free energies using enveloping distribution sampling and one-step perturbation
• Woody Sherman (Schrodinger)
• Harry Stern (Rochester) - Conformational restriction, multiple protonation states, and polarization in protein-ligand binding
• Bill Swope (IBM Almaden Research) [one IBM speaker can be downgraded to 10 min]
• Jeff Wereszczynski (UCSD) - Using Accelerated Molecular Dynamics to Enhance Free Energy Calculations
• Tom Woolf (Johns Hopkins) - Finding, Binding and Analyzing Transition Intermediates: How we might define new drug targets
• Huafeng Xu (DESRES) - Reducing statistical errors in free energy calculations
• Wei Yang (FSU) - High-Order Generalized Ensemble Theory Enables the Sampling of Long-Timescale Events for Free Energy Simulations
• Daniel Zuckerman (University of Pittsburg) - Judging biomolecular sampling, and achieving it with variable-resolution models

10 minutes

• David Case (Rutgers) - Free energy methods in Amber and DOCK
• Christopher Chipot (Nancy Université/UIUC) - still uncertain of travel plans
• Fernando Escobedo (Cornell) - On the coupling of simulations of free energy and transitional kinetics
• Gerhard König (Universität Wien) - Comments on Free Energy Methods: Choosing one that fits your needs
• David Minh (ANL) - Absolute Partition Functions without a Reference State
• John Van Drie (Van Drie Research LLC) - How can imperfect free-energy methods best impact drug discovery?
• Marcus Weber, Zuse Institute Berlin - Simulation of binding kinetics using a transfer operator approach

Session chairs

Note that we will need more session chairs / people to give introductory talks for sessions!

• Kim Branson (Vertex) - session chair
• John Chodera (UCB) - session chair
• David Mobley (UNO) - session chair
• Vijay Pande (Stanford) - session chair
• Michael Shirts (UVa) - session chair

Posters

• Gabriel Rocklin (UCSF)
• Michael Schnieders (Stanford) - Calculation of the thermodynamic properties of crystals (solubility and possibly polymorphs)
• Chris Neale (U Toronto)
• David Minh (ANL) - Absolute Partition Functions without a Reference State
• Jeff Wiseman (Pharmatrope) - Fast free energy calculation based on Monte Carlo sampling of pose and conformation with a flexible protein model
• Mikolai Fajer (UCSD) - may bring poster
• Morgan Lawrenz (UCSD) - Impact of calcium on N1 influenza neuraminidase dynamics and binding free energy
• Yi Wang (UCSD)

Talk titles and abstracts

Huafeng Xu, DE Shaw Research

Reducing statistical errors in free energy calculations

In computing the free energy difference between two thermodynamic states, it is common to introduce a sequence of intermediate states, such that adjacent states overlap significantly in phase space. The statistical error in the estimated free energy difference arises from two sources: inadequate sampling in any of the states, and insufficient overlap between adjacent states. The sampling problem is often alleviated by performing replica exchange simulations, in which adjacent states, or replicas, exchange configurations. I will address the question of how to select the intermediate states so as to minimize the statistical error in the estimated free energy difference, and to maximize the average acceptance probabilities in the replica exchange simulations. We have derived bounds for these quantities in terms of the thermodynamic distance between the intermediates, and our results show that in both cases the intermediates should be chosen as equidistant points along a geodesic connecting the end states.

---

Marcus Weber, Zuse Institute of Berlin

Simulation of binding kinetics using a transfer operator approach

I will briefly present a method for the simulation of binding kinetics on the basis of a transfer operator approach. The talk will mainly focus in the projection of a continuous transfer operator to a low dimensional space of metastable conformations.

---

Pengyu Ren, UT Austin

Free Energy Calculations using Polarizable Multipole-based AMOEBA Potential

We will present recent development of AMOEBA force field, especially the automated parameterization for small ligands. Applications of AMOEBA to solvation and binding free energy calculations with explicit and implicit solvent approaches will be discussed.

---

Morgan Lawrenz, UCSD

Impact of calcium on N1 influenza neuraminidase dynamics and binding free energy

The highly pathogenic influenza strains H5N1 and H1N1 are currently treated with inhibitors of the viral surface protein neuraminidase (N1). Crystal structures of N1 indicate a conserved, high affinity calcium binding site located near the active site. The specific role of this calcium in the enzyme mechanism is unknown, though it has been shown to be important for enzymatic activity and thermostability. We report molecular dynamics (MD) simulations of calcium-bound and calcium-free N1 complexes with the inhibitor oseltamivir (marketed as the drug Tamiflu), independently using both the AMBER FF99SB and GROMOS96 force fields, to give structural insight into calcium stabilization of key framework residues. Y347, which demonstrates similar sampling patterns in the simulations of both force fields, is implicated as an important N1 residue that can "clamp" the ligand into a favorable binding pose. Free energy perturbation and thermodynamic integration calculations, using two different force fields, support the importance of Y347 and indicate a +3 to +5 kcal/mol change in the binding free energy of oseltamivir in the absence of calcium. With the important role of structure-based drug design for neuraminidase inhibitors and the growing literature on emerging strains and subtypes, inclusion of this calcium for active site stability is particularly crucial for computational efforts such as homology modeling, virtual screening, and free energy methods.

---

Emilio Gallicchio, Rutgers

The Binding Energy Distribution Analysis Method (BEDAM): Theory and Applications

The probability distribution of protein-ligand binding energies from the ensemble in which the ligand, while residing in the binding site region, does not interact with the receptor, encodes all of the information necessary to compute the binding constant. I will briefly review the theory behind this formulation, which we named the "Binding Energy Distribution Analysis Method" (BEDAM). Our current implementation of the method is based on implicit solvation, parallel Hamiltonian replica exchange conformational sampling and reweighting techniques. I will present results for the T4 lysozyme system, discuss the challenges posed by targets of pharmaceutical interest, and describe some of the ongoing work to address these challenges.

---

John H Van Drie, Van Drie Research LLC

How can imperfect free-energy methods best impact drug discovery?

The ideal free-energy method would compute an accurate delta-G of binding for an arbitrary ligand, knowing only the apo Xray structure, at an accuracy equivalent to that of an experimental delta-G, in a negligible amount of elapsed time on the computing resources available. Today, we are far short of that ideal, and will likely remain so for some time. Given

(1) the evolving state-of-the-art of free-energy methods, and

(2) knowing that making molecules is rarely the rate-limiting step in a team’s ability to make progress (a team of chemists can bang out a lot of analogs in one week), and

(3) the inability to optimize potency is rarely the reason that a team’s progress “hits the wall”,

where should we attempt to insert these methods into the discovery process to best make an impact? My talk will give my opinions on the answer to this question.

---

David Case, Rutgers

Free energy methods in Amber and DOCK

I will discuss recent developments in implementing various free energy methods in the latest Amber and UCSF-DOCK program suites.

---

Wei Yang, Florida State University

High-Order Generalized Ensemble Theory Enables the Sampling of Long-Timescale Events for Free Energy Simulations

---

Daniel Zuckerman, University of Pittsburgh

Judging biomolecular sampling, and achieving it with variable-resolution models

Every molecular simulation requires good sampling, including free energy calculations where sampling is typically the bottleneck. We have a developed a reasonably simple and general 'yardstick' for measuring equilibrium sampling quality. It applies to both dynamical and non-dynamical methods. Once sampling can be judged, and if fluctuations do indeed matter, it is natural to ask, "What is the most chemically accurate biomolecular model which can be sampled fully?" The answer will vary by problem and resources, and we have therefore developed a flexible platform for variable-resolution modeling. Our memory-based software tracks all atoms at minimal run-time cost, and specific interactions can be chosen as atomistic or simplified. We also show how 'simplified' interactions can be made considerably more sophisticated than standard bead models.

---

Matthew Clark and Jeffrey S. Wiseman Pharmatrope, Ltd

Fast free energy calculation based on Monte Carlo sampling of pose and conformation with a flexible protein model

We have implemented a calculation of ligand-protein binding free energy that integrates free energy by random sampling of ligand poses, conformations and local protein motions as well as interaction with explicit water in the binding site. The method is fast, requiring 3 cpu-hr or less for a ligand with 5-7 rotatable bonds, and requires very little set-up time, needing only a protein model and a ligand structure file as input. The easy setup enables computation of binding free energy for hundreds of ligands, providing the unique ability to examine the performance of free energy computations for diverse molecules. The system was benchmarked on datasets comprising approximately 300 ligands across 3 proteins - T4 lysozyme, p38 kinase, and factor Xa.

---

David D. L. Minh, Argonne National Laboratory

Absolute Partition Functions without a Reference State

I will describe new algorithms for computing absolute partition functions that do not require a reference state. Instead, simple mathematical relationships involving transition functions (conditional probability distributions such as those frequently used in Markov Chain Monte Carlo simulations) are utilized. Promising results with a 2D Ising model hints at potential with more complex systems such as biomolecules.

---

Harry Stern, University of Rochester

Conformational restriction, multiple protonation states, and polarization in protein-ligand binding

We present the application of some new (and old) methods towards addressing three problems in estimating binding affinities. The first is accounting for the free energy cost of ligand conformational restriction upon binding, which we do with a free energy calculation on the ligand using free-energy perturbation from a quasiharmonic reference state. The second is accounting for changes in protonation state; here we use a simple and general approach based on linked binding and protonation-state equilibria. Finally, we examine the importance of electrostatic polarization in binding and present some preliminary work on a fluctuating-charge force field parameterized from density functional theory and crystal structures. Methods are examined by performing calculations for a set of about two hundred protein-ligand complexes, taken from the PDBbind database, for which experimental structures and affinities are both available.

---

Hideaki Fujitani, University of Tokyo

MP-CAFEE: Principles and challenges

Non-equilibrium equality for free energy differences proved by Jarzynski in 1997 is one of the basic physical theories which includes the traditional FEP (free energy perturbation) as a simple case. The binding free energies for proteins and ligands are accurately calculated by MP-CAFEE (massively parallel computation of absolute binding free energy with well-equilibrated states) when we use improved force field parameters. We show some MP-CAFEE results to explain the principles and talk about our new challenges.

---

Jeff Wereszczynski, USCD

Using Accelerated Molecular Dynamics to Enhance Free Energy Calculations

Molecular dynamics (MD) simulations have become a vital tool in computer-aided drug design. Recently, our group has developed a novel methodology for improving the efficiency of MD simulations by altering the underlying potential energy surface to increase conformational sampling. In accelerated molecular dynamics (AMD) underlying thermodynamic quantities pertaining to the un-accelerated, classical system may in theory be recovered through trajectory reweighting. In practice however, accurate statistics are difficult to obtain for protein-sized systems due to a large variance in the weights, which tend to strongly favor only a few conformations. Here, we show that by selectively accelerating a region of interest (for example, a protein’s active site) accurate thermodynamic statistics may be recovered at a reduced computational cost compared to classical MD. We demonstrate this by using free energy perturbation (FEP) with the Bennett acceptance ratio to calculate the binding energy of the influenza drug oseltamivir (marketed as Tamiflu) to the N1 neuraminidase protein. We observe an increase in the efficiency of FEP calculations of two to three times the classical FEP results. These results suggest that AMD may be used to reduce the cost of free energy calculations in methods such as FEP, thermodynamic integration, and envelope distribution sampling.

---

Fernando A. Escobedo

On the coupling of simulations of free energy and transitional kinetics

Typically, simulation methods have been designed to either simulate free energy data or kinetics information of biomolecular processes. Often times, however, free energy data are instrumental to guide the generation and interpretation of kinetic data. Likewise, data on the kinetic of transition pathways can help identify, e.g., suitable reaction coordinates to guide free-energy calculations. More generally, to properly understand and predict the behavior of a system (e.g., when hysteresis or metastable states are important) and mechanism of a process, both types of information are essential. While some methods have already been proposed that can generate transition rates and pathways together with free energies within a common framework, much remains to be done to fully test and optimize them. In this talk, I will review some recent methodological developments that address selected aspects of this coupling of thermodynamic and kinetic data (in the context of simple examples) and underlie some of the outstanding challenges and opportunities.

---

Regis Pomes, Hospital for Sick Children / University of Toronto

Efficient Algorithms for Conformational Sampling in Free Energy Simulations

Conformational sampling of the rugged energy landscapes governing protein dynamics remains one of the essential challenges impeding free energy simulations. A consequence of ruggedness is that restraints or biases must be employed to overcome barriers along the reaction coordinate used in (protein-ligand) binding free energy calculations. However, another consequence of ruggedness is that the use of restraints along the reaction coordinate further impedes conformational isomerization in other degrees of freedom, such as reorientation of the ligand or rotameric transitions of side chains in the active site of an enzyme, leading to an increase in systematic sampling error. A way to address this problem is to use generalized ensemble algorithms such as replica exchange or simulated tempering to attain a random walk of the solute/ligand along the reaction coordinate.

I will present two new generalized ensemble methods, distributed replica sampling and virtual replica exchange, aimed at overcoming the practical limitations of conventional generalized-ensemble simulations. Both methods remove the need for synchronization of multiple copies of the system inherent in replica exchange approaches. In addition, the new methods also avoid the need for extensive initial simulations for the calculation of weight factors or potential energy distribution functions required, respectively, by simulated tempering and serial replica exchange algorithms. The performance of the methods will be examined and discussed for challenging simulations of disordered protein states and binding of peptides and proteins to lipids.

---

Scott P. Brown, Abbott Laboratories

What does it take to positively impact industrial drug design?

While the use of computational models in drug discovery has become an integral part of pharmaceutical research, the application of these models often fails to reach its full potential. This talk will present an analysis of historical data acquired in the course of performing drug research at Abbott Laboratories. By inspecting past performance of medicinal-chemistry campaigns we attempt to quantify the scope of the problem, and propose scenarios in which computational models can expect to have significant influence on drug discovery research.

---

Gerhard König, Universität Wien

Comments on Free Energy Methods: Choosing one that fits your needs

In this talk we attempt to stimulate discussions on the suitability of different equilibrium free energy methods for determining free energy differences in a semi-automated way (as required for drug design), focusing in particular on Thermodynamic Integration (TI) and Bennett's acceptance ratio method (BAR). Recent experiments concerning the efficiency, accuracy and precision of TI and BAR indicate that there is no clear winner. Instead, the optimal choice of carrying out computational free energy experiments depends on various factors, including not only the complexity of the molecular system under study, but also the structure of the computer resources available. To illustrate this, we will briefly evaluate the individual strengths of each method on multiple levels, including computational efficiency (e.g., the minimal number of intermediate steps) and hardware requirements.

---

Huafeng Xu, DESRES

Reducing statistical errors in free energy calculations

In computing the free energy difference between two thermodynamic states, it is common to introduce a sequence of intermediate states, such that adjacent states overlap significantly in phase space. The statistical error in the estimated free energy difference arises from two sources: inadequate sampling in any of the states, and insufficient overlap between adjacent states. The sampling problem is often alleviated by performing replica exchange simulations, in which adjacent states, or replicas, exchange configurations. I will address the question of how to select the intermediate states so as to minimize the statistical error in the estimated free energy difference, and to maximize the average acceptance probabilities in the replica exchange simulations. We have derived bounds for these quantities in terms of the thermodynamic distance between the intermediates, and our results show that in both cases the intermediates should be chosen as equidistant points along a geodesic connecting the end states.

---

William Jorgensen, Yale University

Successes and Challenges for FEP-Guided Optimization of Enzyme Inhibitors

Monte Carlo/FEP calculations for optimization of substituents on an aromatic ring and for choice of heterocycles are now common. Successful application has been achieved for HIV reverse transcriptase, FGFR1 kinase, and macrophage migration inhibitory factor (MIF); micromolar leads have been rapidly advanced to extraordinarily potent inhibitors.

---

Julien Michel, Yale University

Water in binding free energy calculations

The binding of a ligand to a protein generally occurs in aqueous solution and water profoundly influences the nature and energetics of protein-ligand interactions. Computer models that simulate ligand binding must correctly capture the influence of water to represent protein-ligand interactions accurately.

I will present popular strategies to model water in the context of binding free energy calculations, ranging from simple implicit solvent models to detailed explicit solvent models. For each approach I will discuss strengths, pitfalls and possible solutions. I will conclude with an outline of the challenges that remain to be addressed to model solvation in ligand binding using free energy methods.

---

Sereina Riniker

Wilfred F. van Gunsteren Research Group

Laboratory of Physical Chemistry

ETH Zurich, Switzerland

Efficient computation of relative free energies using enveloping distribution sampling and one-step perturbation

The computation of relative free energies of many systems using the thermodynamic integration (TI) methodology and the coupling parameter technique that defines the pathways connecting the various systems is a rather expensive computational affair due to the necessary sampling of many intermediate states along the pathways. Recently, a method that avoids the definition of pathways, enveloping distribution sampling (EDS), has been proposed and shown to be much more efficient than TI. Yet, it still may require one simulation per pair of systems for which the relative free energy is to be computed. If the differences between various systems is not too large, the one-step perturbation (OSP) technique is able to produce all relative free energies from just one simulation of a judiciously chosen reference state. The power and the limitations of the EDS and OSP methods will be discussed using practical examples of molecular systems.

Poster titles and abstracts

Gabriel Rocklin, Sarah Boyce, Brian Shoichet, Ken Dill Department of Pharmaceutical Chemistry, University of California-San Francisco, San Francisco, CA 94158-2518, USA

Absolute binding free energy calculations in a charged, partially solvent-exposed model system: Cytochrome C Peroxidase W191G-Gateless

Predicting absolute binding free energies of charged compounds is a significant challenge for any computational method, as is predicting binding affinities of compounds with different net charges that bind the same protein. The Cytochrome C Peroxidase (CCP) W191G-Gateless model system binds both +1 and neutral compounds, and can be useful for testing computational methods prospectively because new and diverse ligands are relatively easy to discover. The 'Gateless' mutant differs from CCP W191G by a three-residue truncation which partially exposes the binding site to solvent, creating a more 'realistic' binding site and also allowing larger ligands to bind. We are using alchemical free energy calculations to predict the binding affinities of new compounds to this protein to test how alchemical calculations perform in a charged environment and with ligands of different net charge. We will follow up these predictions by measuring the binding affinities by ITC and solving the complex structures by crystallography.