Aggiornamento: 23 gen
Computational analyses and simulations of mathematical models of biochemical networks allow for the comprehension of the underlying behaviour of the cellular systems.
Mathematical models of biochemical networks and in silico analysis of these models are widely known as complementary tools to biological laboratory methods for the comprehension of the mechanisms at the basis of cellular processes, as well as for the understanding of emergent behaviours of cellular processes in both physiological and perturbed conditions. In addition, they can incredibly facilitate the formulation of novel hypotheses that can then be tested with targeted laboratory experiments. However, the simulation of large-scale models, characterised by hundreds or thousands of molecular species and their intermediate complexes, as well as hundreds or thousands of chemical reactions, can easily overtake the capabilities of Central Processing Units (CPUs), making the modelling of biochemical networks a worthless or ineffective effort. In addition, the computational analysis of a model generally requires a huge number of simulations for its calibration or to test the effect of different perturbations. In order to overcome these problems, different High-Performance Computing solutions can be adopted. Among them, Graphics Processing Units (GPUs) showed outstanding performance when used for the parallelization of computational methods in Systems Biology, Computational Biology and Bioinformatics.
“Overall, these analyses demonstrate the relevance of GPU-powered deterministic 171 simulators, especially when high numbers of simulations must be performed or 172 large-scale RBMs are taken into account.” [Tangherloni et al., bioRxiv 2020]
The representation of a biological system by means of a mathematical formulation should consider the scale of the modelled system, the nature of its components, and the possible role played by the biological noise. In such a context, deterministic, stochastic, or hybrid algorithms are used to determine the quantitative variation of molecular species amount in time and/or in space.
When the concentrations of molecular species are high and the effect of biological noise can be neglected, Ordinary Differential Equations (ODEs) are used to model the cellular processes. Given the initial concentrations of the system and the set of kinetic parameters, the temporal dynamics of the system can be simulated by solving the system of ODEs by means of some numerical integrators.
Even though the existing CPU-based ODE solvers are robust and optimised, the simulation of large-scale systems (i.e., a large number of reactions and molecular species) of ODEs can become excessively burdensome. To tackle this problem, we proposed LASSIE (LArge-Scale SImulator), a GPU-accelerated tool tailored for large-scale reaction-based models (RBMs). LASSIE exploits a fine-grained parallelisation strategy to distribute on the GPU cores all the calculations required to solve a system of ODEs.
More information about LASSIE is available on GitLab.
We also developed FiCoS (Fine- and Coarse-grained Simulator), an advanced GPU-powered deterministic simulator, designed to specifically deal with several simulations of middle- and large-scale RBMs. FiCoS effectively realises both a fine- and a coarse-grained parallelization on GPUs. Thanks to its peculiar and optimised design, FiCoS can effectively and efficiently perform hundreds or thousands of simulations of the given RBM by exploiting the power of the modern GPUs.