The Automated force field Topology Builder (ATB, http://compbio.biosci.uq.edu.au/atb ) is a Web-accessible server that can provide topologies and parameters for a wide range of molecules appropriate for use in molecular simulations, computational drug design, and X-ray refinement. The ATB has three primary functions: (1) to act as a repository for molecules that have been parametrized as part of the GROMOS family of force fields, (2) to act as a repository for pre-equilibrated systems for use as starting configurations in molecular dynamics simulations (solvent mixtures, lipid systems pre-equilibrated to adopt a specific phase, etc.), and (3) to generate force field descriptions of novel molecules compatible with the GROMOS family of force fields in a variety of formats (GROMOS, GROMACS, and CNS). Force field descriptions of novel molecules are derived using a multistep process in which results from quantum mechanical (QM) calculations are combined with a knowledge-based approach to ensure compatibility (as far as possible) with a specific parameter set of the GROMOS force field. The ATB has several unique features: (1) It requires that the user stipulate the protonation and tautomeric states of the molecule. (2) The symmetry of the molecule is analyzed to ensure that equivalent atoms are assigned identical parameters. (3) Charge groups are assigned automatically. (4) Where the assignment of a given parameter is ambiguous, a range of possible alternatives is provided. The ATB also provides several validation tools to assist the user to assess the degree to which the topology generated may be appropriate for a given task. In addition to detailing the steps involved in generating a force field topology compatible with a specific GROMOS parameter set (GROMOS 53A6), the challenges involved in the automatic generation of force field parameters for atomic simulations in general are discussed.

To test and validate the Automated force field Topology Builder and Repository (ATB; http://compbio.biosci.uq.edu.au/atb/ ) the hydration free enthalpies for a set of 214 drug-like molecules, including 47 molecules that form part of the SAMPL4 challenge have been estimated using thermodynamic integration and compared to experiment. The calculations were performed using a fully automated protocol that incorporated a dynamic analysis of the convergence and integration error in the selection of intermediate points. The system has been designed and implemented such that hydration free enthalpies can be obtained without manual intervention following the submission of a molecule to the ATB. The overall average unsigned error (AUE) using ATB 2.0 topologies for the complete set of 214 molecules was 6.7 kJ/mol and for molecules within the SAMPL4 7.5 kJ/mol. The root mean square error (RMSE) was 9.5 and 10.0 kJ/mol respectively. However, for molecules containing functional groups that form part of the main GROMOS force field the AUE was 3.4 kJ/mol and the RMSE was 4.0 kJ/mol. This suggests it will be possible to further refine the parameters provided by the ATB based on hydration free enthalpies.

The ability of atomic interaction parameters generated using the Automated Topology Builder and Repository version 3.0 (ATB3.0) to predict experimental hydration free enthalpies (DG water) and solvation free enthalpies in the apolar solvent hexane (DG hexane) is presented. For a validation set of 685 molecules the average unsigned error (AUE) between DG water values calculated using the ATB3.0 and experiment is 3.9 kJ•mol-1. The slope of the line of best fit is 1.00, the intercept-1.0 kJ•mol-1 and the R 2 0.90. For the more restricted set of 239 molecules used to validate OPLS3 (J. Chem. Theory Comput. 2016, 12, 281-296.) the AUE using the ATB3.0 is just 2.7 kJ•mol-1 and the R 2 0.93. A roadmap for further improvement of the ATB parameters is presented together with a discussion of the challenges of validating force fields against the available experimental data.

To enhance efficiency in molecular dynamics simulations, forces that vary slowly are often evaluated less often than those that vary rapidly. We show that the multiple-time-step algorithm implemented in recent versions of GROMACS results in significant differences in the collective properties of a system under conditions where the system was previously stable. The implications of changing the simulation algorithm without assessment of potential artifacts on the parametrization and transferability of effective force fields are discussed.