Dynamic load balancing enables large-scale flux variability analysis

Abstract

Genome-scale metabolic models (GSMMs) of living organisms are used in a wide variety of applications pertaining to health and bioengineering. They are formulated as linear programs (LP) that are often under-determined. Flux Variability Analysis (FVA) characterizes the alternate optimal solution (AOS) space enabling thereby the assessment of the robustness of the solution. fastFVA (FFVA), the C implementation of MATLAB FVA, allowed to gain substantial speed up, although, the parallelism was managed through MATLAB. Here veryfastFVA (VFFVA) is presented, which is a pure C implementation of FVA, that relies on lower level management of parallelism through a hybrid MPI/OpenMP. The flexibility of VFFVA allowed to gain a threefold speedup factor and to decrease memory usage 14 fold in comparison to FFVA. Finally, VFFVA allows processing a higher number of GSMMs in faster times accelerating thereby biomedical modeling and simulation. VFFVA is available online at https://github.com/marouenbg/VFFVA.