A Computational Method to Construct an Extensive Metabolic Pathway Database

Computational metabolic design has become essential in expanding metabolic pathway knowledge from an increasing number of metabolic pathway data. In order to find potential metabolic pathways of high value, it is necessary to expand knowledge from existing databases such as KEGG and BRENDA database by mining comprehensive chemical and enzymatic information. Several computational methods have been already reported for metabolic pathway design, but most cases set a limit on the diversity of chemical and enzymatic information because of a trade-off between the amount of information and the computational feasibility. We have developed an efficient method handling such data to design metabolic pathways including putative compounds and enzymatic reactions.

We have taken advantage of chemical structures and structural differences that include sufficient knowledge for reconstructing metabolic pathways. We first decomposed chemical structure information into atom and bond types to represent chemical structures as feature vectors. As a chemical information resource, a chemical database was constructed from PubChem, one of the largest chemical databases, to ensure the diversity of chemical information. The enzymatic reactions derived from KEGG and BRENDA databases were subsequently defined by the differences between substrate and product feature vectors, and linked with enzymatic information. The simple method in representing chemical structures and reactions allows us to design metabolic pathways on the basis of feature vectors.

The design of metabolic pathways is started by finding combinations of reaction features that satisfy differences between two chemical features. The reaction features are then rearranged to yield chemical features in sequence, which are used to assign compounds from our chemical database by similarity comparison. Compared with the previous method, our method significantly reduces the computational time to find extensive metabolic pathways. The resulting metabolic pathways including putative compounds and enzymatic reactions are ranked on the basis of feasibility criteria using chemical similarity and stored in a pathway database. A web user interface is also developed to check pathway candidates by eye inspections.

As a test case, a set of more than 7,000 alpha amino acid-like compounds from the PubChem database is calculated to find acceptable metabolic pathways for their synthesis from glucose. We found putative metabolic pathways for 1987 compounds, and checked to obtain some interesting pathways out of them. Any chemical and reaction information can be also applied to develop an extensive pathway database, which will increase a chance to find unknown metabolic pathways with diverse chemical and enzymatic information.