Divide-and-Conquer Approach to the Parallel Computation of Elementary Flux Modes in Metabolic Networks

Elementary flux modes are an important class of metabolic pathways used to characterize the functioning and behavior of metabolic networks of biochemical reactions in a biological cell. The computation of the elementary flux modes is accomplished by using the so-called Nullspace Algorithm whose high computational cost and memory requirements still limit the computation to relatively small metabolic networks. We combine a "combinatorial\´\´ parallelization with a novel divide-and-conquer paradigm into a new implementation of the Nullspace Algorithm with lower memory requirements. We discuss the disadvantages of the combinatorial parallelization and divide-and-conquer ideas and explain why their combination attains more computational power. The improved parallel Nullspace Algorithm is used to compute up to nearly 50 million elementary flux modes for a metabolic network for yeast, a task which was previously not possible using either of the two approaches individually.