Multiobjective optimization to reconstruct biological networks

Abstract

Automated methods for reconstructing biological networks are becoming increasingly important in computational systems biology. Public databases containing information on biological processes for hundreds of organisms are assisting in the inference of such networks. This paper proposes a multiobjective genetic algorithm method to reconstruct networks related to metabolism and protein interaction. Such a method utilizes structural properties of scale-free networks and known biological information about individual genes and proteins to reconstruct metabolic networks represented as enzyme graph and protein interaction networks. We test our method on four commonly-used protein networks in yeast. Two are networks related to the metabolism of the yeast: KEGG and BioCyc. The other two datasets are networks from protein-protein interaction: Krogan and BioGrid. Experimental results show that the proposed method is capable of reconstructing biological networks by combining different omics data and structural characteristics of scale-free networks. However, the proposed method to reconstruct the network is time-consuming because several evaluations must be performed. We parallelized this method on GPU to overcome this limitation by parallelizing the objective functions of the presented method. The parallel method shows a significant reduction in the execution time over the GPU card which yields a 492-fold speedup.

Keywords: Biological network reconstruction; Metaheuristic; Optimization; Topological properties.