Dr. Repasky is product manager for a number of Schrödinger programs, including Glide and MCPRO+. In this article, Dr. Repasky discusses some of the features unique to MCPRO+, Schrödinger’s recently released program for predicting ligand binding affinities using FEP methods.
When taking a ‘hit’ compound from high-throughput screening or other discovery approaches to a suitable ‘lead,’ researchers frequently need to rank order compounds by predicted relative or absolute binding affinities. Methods such as MM-GBSA/PBSA and the various flavors of Linear Response methodology have shown considerable success, though they lack a solid theoretical framework. These methods approximate an exact solution that is available using Free Energy Perturbation (FEP) theory. When paired with Monte Carlo sampling and explicit representation of solvent, FEP methods are among the most accurate available means of generating relative free energy differences for use in rank ordering compounds by binding affinity.
Historically, the setup, execution, and analysis of FEP simulations required expert knowledge and considerable time. While developing MCPRO+, we attempted to address these limitations and thereby make accurate rank ordering of compounds by FEP or linear response methodologies broadly accessible to users with a range of experience.
Another goal of MCPRO+ is to generate results in a timely fashion suitable in support of lead optimization projects.
The MCPRO+ simulation package is built upon the MCPRO application from Professor Bill Jorgensen’s laboratory. MCPRO was designed to perform Monte Carlo statistical mechanics simulations and molecular mechanics energy minimizations for biomolecular systems using explicit solvent. MCPRO is well validated and has been applied in many published studies including the calculation of relative and absolute binding affinities, free energies of hydration, linear response predictions of binding affinities, and potentials of mean force predictions.1
To address the issue of complex simulation setup, we have automated common tasks such as the assignment of OPLS force field parameters, the creation of structural topologies to be perturbed as part of an FEP simulation, and the generation of structural inputs that specify the degrees of freedom to be varied throughout the Monte Carlo simulation.
Wizard-based graphical user interfaces have been developed within Maestro that guide users through the process of setting up the most common types of simulations. Using the provided GUIs one can perform Monte Carlo sampling, calculate a single relative free energy difference, complete a free energy cycle required to compute relative free energies of binding, perform energy minimizations, and execute linear response calculations.
To simplify the process of calculating relative binding affinities and to facilitate the study of structure-activity relationships, we have created an interface within Maestro that enables rank ordering of a provided diverse library of fragments at a single ligand substitution site within a protein-ligand complex. This approach allows researchers to determine how binding energy is affected by substituent size, polarity, and specific interactions with the receptor.
While free energy perturbation simulations require significant computation resources, this need has become less onerous thanks to improvements in CPU capabilities and the incorporation of multiple CPU cores in a single processor. FEP simulations are embarrassingly parallel, with each simulation broken into a number of smaller units that can be simultaneously distributed across a large number of processors. To distribute simulations across multiple hosts or a cluster, the user need only execute a single command with one or two input files in a keyword=value format. The simplicity of the input file format allows researchers to easily repeat or modify calculations if need be.
Additionally, real-time progress monitoring tools, both numeric and graphical, provide users with a snapshot of the trends in macroscopic properties.
Finally, we have implemented a sophisticated analysis panel that allows users to import the results or current state of a simulation and create plots for about fifteen properties of interest. Data is plotted as a function of the length of a simulation or number of perturbation windows. With the aid of these analysis tools, it is often straightforward to uncover the factors responsible for observed free energy differences, such as differential protein-ligand interactions and differences in ligand hydration. Deeper understanding of the physical basis for high binding affinity can lead to rapid progress in lead optimization.
Having seen a successful first release as part of the 2007 Schrödinger Suite, the MCPRO+ development team is now taking the first steps toward a second release as part of the planned 2008 Schrödinger suite. Among the enhancements we’re considering are improved sampling efficiency for protein-ligand complexes, simplified examination of protein mutation profiles (also called resistance profiles), a new tool for determining the optimum heterocyclic substituents for maximum binding in ring systems, further speed improvements, and additional refinements to the MCPRO+ workflow.
1 Jorgensen, W. L., Tirado-Rives, J., "Molecular Modeling of Organic and Biomolecular Systems using BOSS and MCPRO", J. Comp. Chem. 2005, 26(16), 1689-1700.