With the emergence of the Exascale Era in High Performance Computing, unprecedented biological phenomena can be explored. Whole viruses and large macromolecular complexes are being modeled atomistically [1] [4] [5] [6] [8] [14]. Molecular level simulations can test previous models of biomechanical continuum models of important biological processes such as muscle contraction and transport through the nuclear pore complex. Simulations can produce accurate free energy profiles that determine ion selectivity and permeation through channel proteins. Large scale models of molecular binding to proteins or nucleic acids can help select viable drug candidates. Advanced enhanced sampling simulations can describe quantitatively the free energy landscape and the kinetics [2] of complex and lengthy biological processes.

Numerous codes, typically written in C++ or Fortran, have been developed for biomolecular simulations [9] [10]. These codes have been optimized to run on advanced architectures such as GPU accelerators with high efficiency. Unfortunately, however, some of these codes are specialized and they might encounter difficulties to be broadly applicable, adaptable, and ready to tackle grand challenge problems in phenomena that involve many biological partners, such as neuronal signaling or the entry of virus into its target host cell.

An array of innovative, advanced enhanced sampling simulations can describe quantitatively the free energy landscape and the kinetics of complex biological processes [7]. In spite of these excellent results, however, often the algorithms do not scale as efficiently as one would like. Within this scenario, an ongoing revolution in computational science is taking place. This revolution is facilitated by the development of big data and artificial intelligence tools which promise to dramatically enlarge the scope of biomolecular simulation [3] [11] [12] [13]. A pressing challenge is therefore to implement these algorithms with great efficiency into the existing codes.