Elucidating the biochemical network of living and synthetic systems is a challenging problem. We are developing PathMiner, a computational framework for exploring metabolic pathways.
The genomes of several dozen species are publicly available, and databases of gene products and biochemistry are coming online. In our current system, we used the KEGG database, which includes the annotated genomes of over 100 organisms and over 5000 enzyme-catalyzed reactions. From this database of enzymes, we can reconstruct various metabolic networks of interest to microbiologists and metabolic engineers. To make inferences about pathways we abstract metabolism as a state-space in which compounds are points and biotransformations are state-transitions. We represent compounds using a set of chemical descriptors including, atoms, bonds, and molecular graphs. Though the dimensions are customizable, we have found exciting results by just considering the atomic and bond content of the metabolite "states". Additionally, we have also explored deeper graph-based representations of metabolites are transformations, which has allowed us to make inferences about symbolic biochemical rules in enzymatic transformations. Finally, we are able to apply these rules for de novo pathway prediction. Our pathway inference algorithm and biochemical knowledge are encoded in Lisp, which is a flexible symbolic computation environment and is ideally suited for rapid prototyping.
In this poster, we discuss applications of PathMiner to two quite different biological problems.
First, we discuss our use of PathMiner in developing capabilities for whole network analysis and metabolic reconstruction for organisms from genomes. Utilizing the annotated genomes in KEGG and GenBank, we are working on a complete network metabolic reconstruction of the methanogen Methanococcus jannaschii. Because it is an autotroph, the metabolic network of this organism has some unique properties that we exploit in the reconstruction. We utilize the enzymatic prediction capabilities of PathMiner to fill in the metabolic gaps, which cannot be annotated based on sequence.
Second, we are using PathMiner to contributed to metabolic engineering as well. As an example, we used PathMiner to explore the synthesis of 1,3 propanediol from sugar. DuPont and Genencor recently announced a microbial process for the generation of 1,3 propanediol from corn sugar. The technique involves the modification of E.coli with genes from yeast and Klebsiella pneumonae. While the genome for Klebsiella is not in our database, we present some novel alternative possibilities for engineering single microbial synthesis of 1,3 propanediol using an enzyme from the hyperthermophilic bacterium Aquifex aeolicus.