Parallel Simulation Engines for Whole-Cell Models

The enormous complexity of whole-cell models needs new algorithmic approaches and high performant simulators. Our goal is the integration of deterministic and stochastic simulation algorithms in a highly scalable parallel simulator. Our hardware platform is a cluster of workstations as they are available in most research institutions. Therefore our parallel simulation engines are based on the Message Passing Interface (MPI) [1] standard, which has been implemented for many hard- and software environments. MPI is available as a library and represents a reliable platform for our work. We have implemented two parallel simulation engines, the Parallel Deterministic Simulation Engine (PDSE) and the Parallel Stochastic Simulation Engine (PSSE). Both use MPI for interprocess communication and share the same interface. Sharing a single library for parallel processing (MPI) reduces the amount of external software our simulator depends on and the system becomes easier to install and maintain. The PDSE is written in C++ and based on the Portable Extensible Toolkit for Scientific Computing (PETSc) [2], which is a software library providing routines and data structures for the numerical solution of partial differential equations and related problems on parallel (and serial) computers. PETSc uses MPI (as described above) for interprocess communication. The PDSE performs simulations on a distributed (regular) grid, that can be either 2- or 3-dimensional. The size of the grid may be customized according to the reaction volume to be simulated. Using the forward or the backward Euler method, PETSc solves the ordinary differential equations arising from the discretization of time-dependent partial differential equations that describe the biochemical model under investigation. Concerning the solution of ordinary differential equations we can rely on the stable numerical methods provided by the PETSc library. However, we are also able to use our own routines, since PETSc offers an interface to external ODE solvers as well. The PSSE is also written in C++ and uses an efficient implementation of the Next Reaction Method introduced by Gibson [3]. The reaction volume is subdivided into a cubic grid and each sub