Metabolic Network Analysis with MetaboNetworks

First, the appropriate data from KEGG has to be imported to Matlab. In MetaboNetworks this is done using a function that uses the KEGG REST-API to calculate a metabolite adjacency matrix that can later be used to draw the graphs. The user can select one or multiple organisms for which complete genomes are available in KEGG; for these organisms a list of enzymes (with E.C. numbers) that are associated with a gene from any of the organisms is determined. Using this enzyme list, all reactions are queried and enzymes involved in the reactions are matched against the enzyme list. Only reactions that require an enzyme from the list or that are listed as ‘non-enzymatic’ or ‘spontaneous’ are used to find their main reaction pairs. The compounds from these reaction pairs are considered adjacent. Each row/column in the adjacency matrix indicates a specific compound (with a KEGG compound ID) and a list of all names for these compounds are found from the KEGG compound database. A reaction database has previously been collected using a similar approach (Ma and Zeng, 2003 (link)) to MetaboNetworks, however, that database includes reactions from all species, whereas MetaboNetworks focusses on organisms of interest as not all reactions can occur in all organisms.
Second, when the data collection is complete, MetaboNetworks can be used to create and explore custom networks. A list of metabolites, e.g. biomarkers arising from a metabonomic experiment, can be passed to MetaboNetworks and it searches for the shortest path between each of these metabolites using the breadth-first search algorithm. All compounds that are a part of a shortest path between any of the metabolites are included in the network. By default, MetaboNetworks plots the network as a circular graph. Other graph layouts include a spring-embedded layout, high-dimensional embedding and two types of uniform edge-length layouts, the last aim to place nodes with as little overlap as possible. If the Matlab statistical toolbox is installed, multidimensional scaling can also be used.
Last, when the initial network layout is satisfactory the graph layout can be manually adjusted. Supported adjustments include node position, node/edge removal, highlighting nodes (see green edges of nodes in Fig. 1), and shortest paths (orange edges in Fig. 1), node text and nodes/edge/text properties (font, width, size, etc.). If additional data is supplied, the association of the metabolites with a response variable can be shown as node colour (see Fig. 1). Furthermore, the network can be exported as a tif, png, pdf, eps or other image formats, the network can always be reset to the original graph (all changes are lost). Another option is to click on a node to open a web browser showing the compound entry in KEGG or show reactions pairs in KEGG of selected nodes. The Supplementary Information includes a full walkthrough of the software and all the capabilities.
Fig. 1.

Shows the graphical user-interface of MetaboNetworks with a custom network drawn for significant metabolites from a hydrazine toxicity study in rats (Nicholls et al., 2001 (link)). Metabolites higher in hydrazine-dosed rats compared with controls are shown in red, and metabolites lower in hydrazine-dosed rats are shown in blue. The white nodes are part of shortest paths between the coloured nodes. The edges shown in orange are part of the shortest path (four reactions) between taurine and glycine. Aside from the rat, all bacteroidetes and firmicutes species were included in the database