Efficient Searching and Annotation of Metabolic Networks Using Chemical Similarity

The ability to quickly search large biochemical networks to discover or design novel metabolic pathways is urgently needed as metabolic engineers both become more aggressive with biosynthetic goals and have access to exponentially increasing amounts of biochemical data. In particular, novel metabolic routes have the potential to offer higher yields and better economy for pharmaceuticals and fine chemicals. Existing network search and generation tools are powerful, but are also limited by the combinatorial explosion of potential compounds and reactions in pathways that can originate from a single compound node in the biochemical network. Moreover, existing algorithms cannot efficiently annotate reactions with non-native enzyme substrates, despite this being a core activity for selecting enzymes for engineering. To facilitate both speed and novel pathway prediction, two algorithms, SimIndex and SimZyme, have been written to efficiently search novel biochemical networks and to predict particular enzymes thatâ??via enzyme promiscuityâ??can catalyze reactions required in de novo pathways that proceed through novel intermediates.
A marked improvement in network search performance is achieved with SimIndex when it is coupled with the predictive power of the biochemical network generation software BNICE. For test pathways with three and four compounds, SimIndex offers order-of-magnitude reductions in the number of nodes searched while still reaching the desired target molecule. The reduction of both nodes and edges (compounds and reactions, respectively) searched enables faster pathway identification within the putative network. Our method of assigning specific enzymes to putative reactions, SimZyme, ranks enzymes likely to perform a desired chemical transformation based on information in the BRENDA Database. We validated SimZyme using a leave-one-out approach to determine how the algorithm would predict the enzymatic reaction we removed. Across four enzyme classes, the correct enzyme was ranked in the top three results in
80%-95% of the 100 trial runs in each class. Our preliminary results indicate that our approaches to network reduction and edge annotation in biochemical networks are useful strategies for efficient assembly of novel metabolic pathways in silico, which in turn facilitate more efficient translational work in the laboratory.