Fast and Accurate Simulation Framework Targeting Molecular Dynamics for Cells Substructures

Molecular simulations are continuously shedding light into a wide range of biological phenomena such as cell structural stability, intracellular processes, cells interaction with drugs and membrane permeability. Hence, the search for better and faster approaches of molecular simulations to understand these activities related to cells dynamics, is becoming more and more challenging. This thesis firstly proposes a multiscale modeling approach that provides a link between atomistic level and continuum scale modeling concerning subcellular proteins simulations. This methodology represents a first step of a more complex multiscale model able to simulate and predict the effects of specific local changes on the subcellular proteins mechanics due to pathological conditions such as single point mutations, or therapeutic treatments such as the effects of bound pharmacological molecules, and their consequences on the overall properties of the filament. Although this approach takes actin filaments as case study, it devises the guidelines on multiscale modeling of several supramolecular hierarchical assemblies such as microtubules, collagen fibers, and polymers whose localized nanoscale phenomena reflect on a macroscale level of description. Moreover, this thesis presents optimization strategies for molecular simulations, proposing graphics processing units (GPU) as an excellent high performance computing commodity hardware for this purpose, paying particular attention to the accuracy of the simulations. A coarse grain (CG) molecular dynamics (MD) simulator specialized for simulations of lipid bilayers, is considered as a case study and optimized and accelerated for single-GPU environment. A speed-up of 30 fold for water systems and 15 fold for lipids is obtained when exploiting CUDA intrinsic functions for the floating point arithmetic in the GTX480 GPU with respect to CPU simulations. This research performs a detailed analysis of CG features and their impact on the achievable acceleration, to formulate guidelines for writing more efficient CG MD codes. Finally, the CG MD model considered as case study is validated and integrated for multiple-GPU environment and thereafter employed to perform analysis of large-scale systems simulations investigating the phase transition of lipid bilayers of cell membranes in a supercomputing context