Accurate molecular dynamics simulations rely on an underlying potential function that must describe the physical interactions sufficiently well. Accurate treatment of electrostatic interactions in biomolecular simulations has long been recognised as key to modelling many biological processes and thermodynamic properties of interest. Fixed point-charge force fields have been highly successful in predicting simple properties, such as hydration or binding free energies, at least for non-ionic species. Accuracies of better than 1 kcal/mol have been achieved in many cases. This simple electrostatic treatment is much more problematic for ionic groups that interact in heterogeneous environments such as macromolecular interfaces, where electronic polarizability can play an important role.

To address these known limitations, more sophisticated functional forms of the electrostatic interactions have been proposed. These include the use of distributed multipoles to provide a better representation of the molecular electrostatic potential, and models that explicitly represent electronic polarization, either through induced dipoles, Drude pseudo-particles, or fluctuating atomic charges. Beyond polarisable force fields, QM/MM methods with explicit coupling of polarisation between QM and MM regions have also been developed. While these methods have been implemented in several major MD codes, the availability of highly parallel and efficient implementations is very recent and far from complete, both in terms of MD codes, code functionality, system applicability, and algorithmics.

While the improved representation of the underlying physics should of course yield improved accuracy, the use of more elaborate functional forms introduces additional costs in terms of more parameters to optimise, and longer calculation times. Thus, there are still major gaps in what molecules can be described by each polarizable force field, and systematic comparisons between the methods are mostly lacking.

Our 2016 CECAM workshop on this topic examined the state of the art with these new potentials, considered the likely advances over the next few years, and the issues which stand in the way of more widespread adoption of these methods. As we concluded in our 2016 workshop report, it is important, three years later, to reconsider these questions in the light of recent progress. In the last three years, there has been major progress in code efficiency and availability, the first applications (as opposed to benchmarks) of polarizable force fields to proteins, and significant extensions of force fields to new molecules like DNA or ATP.

The main remaining questions are similar to those in 2016:

1. What are the improvements given by including advanced physics and how does parameterisation affect these? Recent applications of AMOEBA to the Sampl competition for binding free energies gave somewhat disappointing results, indicating parameterization is still suboptimal. Recent applications of Drude to protein-peptide binding free energies gave a significant improvement over additive force fields, but the improvement was not uniform even within the same system. Studies evaluating the effects of including multipole descriptions of electrostatics or induced dipoles in protein force fields are just beginning. Numerical analyses of the models have sometimes been impaired by incorrect or suboptimal methods to relate experimental and simulated free energies. The effect of over-simple water models needs more investigation.
2. Which implementations are most efficient and accurate? While the Drude pseudo-particle approach is algorithmically simple, and benefited from the outset from parallel code, other models such as induced dipoles are just now becoming avalable in very fast parallel codes, and algorithmic development is still very active. Continuing implementation and testing of these methods in different software and application environments is very much required.
3. Where are advanced electrostatics necessary to answer questions that existing models cannot? Uptake of new methodologies has been limited, partly because fixed point-charge force fields are seemingly ubiquitously successful. Building on knowledge of where advanced electrostatics treatment is necessary (e.g. surface/interface properties, multivalent ionic interactions) will extend the reach of advanced potential functions to new applications.

Our CECAM workshop will address the known limitations of additive fixed point-charge electrostatics and discuss specifically when and how more sophisticated energy functions are best applied. It will bring together developers of these new physical models and users of the key simulation force fields (AMOEBA, CHARMM-DRUDE, SIBFA, NEMO, GEM, GROMOS) to answer these questions.

The workshop will address:

The deficiencies of conventional fixed-charge force fields. The fixed-charge approximation has proved remarkably robust and successful, largely thanks to the many generations of effort in parameterisation. But it consistently gives significant errors (above 1 kcal/mol) for electrostatic free energies in biomolecules with occasional enormous errors. The purpose of this section is to identify clear cases where fixed-charge models fail and careful reparameterisation will not recover accuracy. [Key contributors: Essex, Simonson, MacKerell, Söderhjelm, van Gunsteren]
The hierarchies of methodologies for moving beyond this approximation, including multipole-based methods and polarisable potentials. There are many ways of improving on the point-charge model, particularly in the force fields named above. The purpose of this section is to examine the available options, and to assess their relative merits in terms of accuracy and associated computational and parameterisation cost. [Key contributors: Meuwly, Piquemal, Brooks, Ponder, Cisneros, Söderhjelm, Barone]
Issues of parameterisation of advanced electrostatic models. This section will address how parameters that are robust, accurate and transferable may be derived, focusing particularly on how compatibility with other components of the underlying force field may be achieved. [Key contributors: Ponder, Ren, Mackerell, Lemkul, Roux, Meuwly]
Issues of implementation of these advanced potential energy functions. Given a particular choice of physical representation of advanced electrostatics, say, for example, the use of inducible dipoles to capture explicit polarisation, the associated equations may be solved using different routes. In this example, simple optimisation or extended Lagrangian methods may be used. The purpose of this section is to examine the merits of alternative approaches, the underlying software currently used to deploy these models, and to focus on particular computational black-spots. [Key contributors: Piquemal, Head-Gordon, Roux, Brooks, Ren, Ponder, Cisneros]
Applications of these methods, in particular the contexts in which the expense of these new force fields is rewarded with improved scientific insight. The widespread adoption of more sophisticated models will require clear exemplar applications demonstrating how the improved accuracy they achieve offsets the associated computational cost. This section will explore existing applications and propose new ideas, with particular reference to the deficiencies of fixed point-charge models discussed in the first session. [Key contributors: Simonson, Essex, MacKerell, Meuwly, Schnieders]

We propose to hold six half-day sessions organised around these topics, with introductory and summary talks. Our intention is to reach conclusions regarding what level of electrostatic model is appropriate for a particular problem, and how it may be best implemented and parameterised. To that end, the identified key workers in the sessions will lead discussion based on their ‘hands-on’ experiences of their own methodology development, applications and results. To allow diverse insights while being small enough to encourage discussion, we anticipate approximately 30 participants to attend.