From DNA to FBA: How to Build Your Own Genome-Scale Metabolic Model

Abstract

Microbiological studies are increasingly relying on in silico methods to perform exploration and rapid analysis of genomic data, and functional genomics studies are supplemented by the new perspectives that genome-scale metabolic models offer. A mathematical model consisting of a microbe's entire metabolic map can be rapidly determined from whole-genome sequencing and annotating the genomic material encoded in its DNA. Flux-balance analysis (FBA), a linear programming technique that uses metabolic models to predict the phenotypic responses imposed by environmental elements and factors, is the leading method to simulate and manipulate cellular growth in silico. However, the process of creating an accurate model to use in FBA consists of a series of steps involving a multitude of connections between bioinformatics databases, enzyme resources, and metabolic pathways. We present the methodology and procedure to obtain a metabolic model using PyFBA, an extensible Python-based open-source software package aimed to provide a platform where functional annotations are used to build metabolic models (http://linsalrob.github.io/PyFBA). Backed by the Model SEED biochemistry database, PyFBA contains methods to reconstruct a microbe's metabolic map, run FBA upon different media conditions, and gap-fill its metabolism. The extensibility of PyFBA facilitates novel techniques in creating accurate genome-scale metabolic models.

Keywords: flux-balance analysis; genome annotation; in silico modeling; metabolic modeling; metabolic reconstruction; model SEED.

FIGURE 1

Roles to complexes to reactions. Functional roles have a many-to-many relationship with enzyme complexes. Similarly, enzyme complexes have a many-to-many relationship with biochemical reactions. (A) A one-to-many relationship from roles to complexes. (B) A many-to-one relationship from roles to complexes. (C) A one-to-many relationship from complexes to reactions. (D) A many-to-one relationship from complexes to reactions.

FIGURE 2

Flux-balance analysis (FBA). (A) Example of a bacterial metabolic model displaying two compartments separated by a dashed boundary (extracellular and cytoplasm), seven reactions labeled in blue text (four intracellular and three transporters), and five compounds. (B) The stoichiometric matrix S with corresponding stoichiometric coefficients, and the flux vector v. Each matrix-cell represents the number of compound molecules required for the particular reaction. The integer sign denotes the compound as a reactant (negative value) or as a product (positive value). A zero means the compound is not involved in the reaction. Reversible reactions are typically present in the matrix in one direction. In the instance that a reaction is reversed in the solution, the metabolic flux value for the corresponding reaction will be negative, thus indicating a switch in directionality. (C) The linear programming problem. The mass balance equations constrain the change of compound concentration over time to zero. The final constraints on the system are the physicochemical enzymatic bounds. These bounds signify directionality (e.g., v₃ has a lower bound of 0 and can only produce compound A, whereas v₂ is completely unbounded and can proceed in both directions). The objective function is indicated to maximize the flux of v₄ (i.e., the production of compound C).

FIGURE 3

Controlling oxygen exchange example. The lower bound in exchange reactions (e.g., EX_O2_e0) moderate the availability of a compound for the extracellular compartment. To simulate an anaerobic environment, flux from the environment to extracellular space is cutoff by setting the lower bound to zero, whereas the lower bound is set to negative infinity for an aerobic environment, allowing unlimited amount of oxygen into the extracellular space.

FIGURE 4

PyFBA workflow. Each step to obtain a genome-scale metabolic model using PyFBA is presented on the right. Arrows represent actions to be taken to obtain the next set of information in the workflow. Corresponding to each step are the necessary PyFBA module functions on the left. The gap-fill step is associated with multiple functions, each one differing by the gap-fill strategy used to identify reactions. The gap-fill method is iterative, hence a subset or all of these modules are used to obtain a growing model. The RAST portion of the workflow is an example for using functional role annotations; these steps can be replaced by other methods yielding the same information.