Flux balance analysis (FBA) predicts the optimal flux distribution through a metabolic network by formulating a linear programming problem: maximize a biological objective function (typically biomass production) subject to stoichiometric constraints (S * v = 0), reaction bounds (capacity limits and reversibility), and measured uptake rates. FBA does not require kinetic parameters, making it applicable to genome-scale metabolic models with thousands of reactions. It successfully predicts growth rates, gene essentiality, and metabolic phenotypes across diverse organisms and conditions, and forms the foundation of the COBRA (Constraint-Based Reconstruction and Analysis) toolbox widely used in systems biology and metabolic engineering.
Stoichiometric modeling establishes the space of all metabolic flux distributions compatible with mass balance. Flux balance analysis asks: which point in that space does the cell actually use? The answer comes from optimization. FBA posits that evolution has selected cells to maximize some objective — most commonly the rate of biomass production (growth rate) — subject to the physicochemical constraints of stoichiometry, thermodynamics, and enzyme capacity.
Mathematically, FBA is a linear program: maximize c^T * v (where c is a vector defining the objective, typically the biomass reaction) subject to S * v = 0 (stoichiometric balance), v_min <= v <= v_max (flux bounds from reversibility, measured uptake rates, and enzyme capacities). Linear programming solvers find the optimal solution efficiently even for systems with thousands of variables, which is why FBA works at genome scale. The biomass reaction itself is a pseudo-reaction that consumes amino acids, nucleotides, lipids, and cofactors in the ratios needed to build new cell mass — essentially encoding the cell's biosynthetic requirements.
FBA's most powerful applications are in gene essentiality prediction and metabolic engineering. To simulate a gene knockout, the corresponding reaction's flux bounds are set to zero and the LP is re-solved. If the optimal biomass flux drops to zero, the gene is predicted essential. Across model organisms, FBA correctly predicts gene essentiality with roughly 90% accuracy — remarkable given that it uses no kinetic parameters. For metabolic engineering, FBA identifies which reactions to overexpress, delete, or introduce to redirect flux toward a desired product. Algorithms like OptKnock systematically search for gene deletion strategies that couple product formation to growth — ensuring the engineered organism must produce the desired compound to survive.
The limitations of FBA are well understood and have motivated extensions. FBA predicts steady-state behavior only — no dynamics. It requires an assumed objective function, which may not apply to all cell types or conditions. The optimal solution is often non-unique (many flux distributions achieve the same maximum growth rate), requiring additional methods like flux variability analysis (FVA) to characterize the range of possible fluxes. Despite these limitations, FBA and the COBRA framework remain the most widely used computational tools in systems and synthetic biology, precisely because they deliver useful predictions from minimal data — stoichiometry and a few measured exchange fluxes.