Professor Friesner is a founder of Schrödinger and Professor of Chemistry and Director of the Center for Biomolecular Simulations at Columbia University. As chairman of Schrödinger’s Scientific Advisory Board, Professor Friesner provides strategic vision and guidance for Schrödinger's scientific advancements. In this first installment of Rich's regular column, he reports on the progress of the force field development efforts.
Force fields are employed by most calculations in which molecular modeling is used to study protein structure and protein-ligand binding. The quality of the results obtained from such calculations is in a fundamental sense limited by the quality of the force field. In many cases, reasonable results are obtained despite significant errors in the force field, due to favorable cancellation of errors. Nevertheless, if computational chemistry is to advance to the point where accurate solutions can be delivered robustly over a wide range of chemistry, advances in force field quality will be required.
We have undertaken a parallel, two-tiered approach to force field development at Schrödinger. In the long term, we believe that qualitative improvements in the electrostatic model of the force field, incorporating higher multipole moments and electrostatic polarization, will be required to achieve the highest possible accuracy. However, force fields based on traditional atom-centered fixed charges will continue to play a major role for the foreseeable future, based on speed, ease of use, and effort required to develop new parameters. Hence, we continue to work on both an improved fixed-charge force field and a next generation polarizable force field. My comments in this column are centered on the fixed-charge force field, a major revision of which is scheduled for Schrödinger’s 2007 software release.
Our fixed-charge force field development efforts begin with the OPLS-AA force field of Jorgensen and coworkers, and are focused in the following areas:
(1) Implementation of a novel AM1-BCC based charge methodology, which effectively combines information from ab initio and semiempirical quantum chemistry, as well as a core of OPLS-AA charges derived from liquid-state simulations
(2) Optimization of van der Waals parameters via liquid state free energy simulations
(3) Calculation of accurate valence parameters (stretches, bends, torsions) with broad coverage of functionalities relevant to medicinal chemistry, using high-level ab initio quantum chemical methodology.
This approach has evolved from interactions between the Jorgensen group, MMFF developer Tom Halgren, and my group at Columbia. Implementation is carried out primarily at Schrödinger, where professional software development and parameterization efforts can be performed efficiently.
We have identified inadequate torsional coverage of medicinal chemistry space as a major problem for all currently available force fields, and perhaps the leading source of error at present. Analysis of a virtual library of one million purchasable drug-like compounds identifies a missing torsion (using OPLS-AA atom quartets) in 50% of all cases when MMFF is used as the force field, which can result in errors as large as 5-10 kcal/mol in relative conformational energies for the torsional degree of freedom in question. While approximately 12,000 parameters would be needed to achieve complete coverage for 99% of the molecules in the virtual library, MMFF has only 500 parameters defined.
Based on this analysis, we have generated quantum chemical data at the LMP2/cc-pVTZ(f) level for 12,000 model molecules to enable accurate generation of corresponding torsional parameters. This was done with the aid of a novel tool we developed, which extracts the smallest possible model compound from a larger molecule with a missing torsional parameter. The protocol for fitting parameters has also been automated, and we expect to offer a package of automated tools that will enable users to augment the force field with new parameters in cases where missing torsional parameters are identified.
We intend to carry out extensive tests of the new force field in a diverse range of applications, including docking and scoring, 3D pharmacophore modeling, protein structure prediction and refinement, and free energy perturbation calculations. Previous work on force field parameters for proteins has shown that fitting torsions to high-level ab initio calculations yields better results in protein structural prediction.
Our expectation is that for the first time, no matter what compound they are considering, modelers can feel confident that the force field will provide at the very least a reasonable representation of the energetics. Such broad coverage of chemical space will particularly benefit industrial applications, where the chemistry to be considered is dictated by the needs of the project, and not the limits of modeling methodology.
Comments and questions on Dr. Friesner’s column are welcome. Please send these via email to ask-rich@schrodinger.com, and we’ll post particularly interesting questions and answers in the Q & A section below.
Q & A - Coming soon