Gap Filling of Metabolic Networks Using Flux Coupling Analysis and Expression Profiles of Metabolic Genes

“Gap filling” is the process of finding missing reactions of a metabolic network. Existing gap filling algorithms have mainly focused on finding blocked reactions, and then, adding a minimal number of reactions to the network to unblock those reactions. In the present work, we introduce a novel approach for gap filling that uses flux coupling analysis together with the publicly available datasets of gene co-expressions. In this approach, we assume that if two fully coupled reactions do not have high values of gene co-expression, then these two reactions are not fully coupled in reality, and hence, there must be a gap in the network. The goal of our method is to complete a metabolic network such that there is maximum consistency between experimental gene expression data and theoretical flux coupling relations. For this purpose, in a two-step algorithm based on a mixed integer linear programming formulation, we try to minimize the discrepancies by adding a minimum number of reactions (from a dataset of all known reactions) to the network. When we compared the results of our method with other popular gap filling methods, we observed that our method is able to predict missing reactions that are not detectable by other methods. This difference is presumably due to the fact that our algorithm looks for gaps that cause incorrect prediction of gene co-expression. Obviously, such gaps may not be related to the blocked reactions of the network.  In conclusion, we can recommend our algorithm to be used as a complementary strategy to the existing gap filling strategies.“Gap filling” is the process of finding missing reactions in a metabolic network. Existing gap filling algorithms have focused mainly on finding blocked reactions and subsequent unblocking by adding a minimal number of reactions to the network. In the present work, we introduce a novel approach for gap filling that uses flux coupling analysis together with the publicly available transcriptome datasets (for computing gene co-expression values). In this approach, we assume that if genes in two seemingly fully coupled reactions are not highly co-expressed, then these two reactions are not fully coupled in reality, and hence, there is a gap in the network. The goal of our method is to complete a metabolic network such that there is maximum consistency between experimental transcriptomic data and theoretical flux coupling relations. For this purpose, in a two-step algorithm based on a mixed integer linear programming formulation, we minimize the discrepancies by adding a minimum number of reactions (from a dataset of all known reactions) to the network. By comparing the results of our analysis with the outcome of other popular gap filling techniques, we observed that the method proposed in this work is able to predict missing reactions that are not detectable by other approaches. This difference is presumably due to the fact that our algorithm specifically looks for gaps that cause incorrect prediction of gene co-expression. As such gaps will not necessarily correspond to the blocked reactions, they may not be detected by earlier approaches.  In conclusion, we propose that our algorithm may be used as a complementary strategy for the gap filling problem of metabolic networks.