MiMiC: Multiscale Modeling in Computational Chemistry

Abstract

Hybrid quantum mechanics/molecular mechanics (QM/MM) approaches are commonly used methods for investigating a plethora of chemical, biochemical, and biophysical processes that require explicit treatment of the electronic degrees of freedom when the system is too big to be entirely treated by QM methods alone (Warshel and Levitt, 1976; Senn and Thiel, 2009; Adhireksan et al., 2014; Campomanes et al., 2014, 2015; Brunk and Rothlisberger, 2015; Genna et al., 2016; Li et al., 2017; Cupellini et al., 2018; Loco et al., 2018; Morzan et al., 2018). It is often the method of choice for computational investigations of systems with more than a few thousand atoms (which is commonly the case for biological systems). In QM/MM, the system is split into two parts: a smaller part that is treated at the QM level of theory, whereas the remainder is described at the MM level, which is a computationally more expedient description. In this way, local electronic effects can be captured with the accuracy of a first-principles method, while at the same time explicitly including the effects of the environment at a reasonable computational cost. Current QM/MM implementations have roughly followed either of two strategies: (1) tight integration of QM and MM modules in a single software package or (2) loose coupling of separate QM and MM codes. Strategy (1) generally profits from computational efficiency due to the ability to pass data between the submodules directly (via function calls) but suffers from limited flexibility, since the available choice of methods is often restricted and extensions to different programs may require formidable programming efforts. In contrast, strategy (2), which is typically implemented resorting to data exchange between QM and MM codes <i>via</i> file input and output, enables high flexibility but penalizes efficiency because of increased communication overhead. However, with the field rapidly growing, new simulation paradigms and approaches might quickly emerge, clearly favoring strategy (2) over (1). In the following, we show that flexibility does not necessarily come at the expense of a high computation (or communication) overhead by presenting the recently developed MiMiC framework (Bolnykh et al., 2019; Olsen et al., 2019) that combines the capability of performing fast and efficient multiscale molecular dynamics (MD) simulations with facile support for flexible extensions. These objectives are achieved by applying (2) with an efficient method to exchange data among the coupled software packages. In practice, MiMiC implements a multiple program-multiple data (MPMD) paradigm through a message passing interface (MPI)-based communication library, which allows the entities collaborating within MiMiC to exchange data efficiently. Overall, MiMiC represents a highly modular and general multiscale simulation framework that enables the combination of multiple resolutions and methods for different parts of a system, while retaining high computational efficiency. Moreover, MiMiC was designed to have a flexible architecture enabling multiple resolutions, implementation of different types of coupling (e.g., QM/QM, QM/QM/MM, etc.), and to straightforwardly incorporate emerging—and future—methods and software packages in the field of computational chemistry. This flexibility is of utmost importance in the light of the rapid development of computational methods enabling researchers to tackle complex scientific problems with more and more degrees of freedom that require the incorporation of multiple space and time resolution scales on the one hand, and the rapid advent of new computational approaches on the other hand.