Network design in cells: Optimization of metabolic networks

The DNA of living organisms is an emerging substrate for engineering. Owing to rapid improvements in technology, the cost of reading and writing DNA will likely be nominal in the near future, enabling easy fabrication of DNA. But the question is, What DNA sequence do we make? Given the vast capabilities of living organisms, from sensing, to signaling and communication, to motility, to chemical synthesis, the possibilities of engineered organisms are enormous. Unfortunately, the mapping from DNA sequence to function is not well understood, and determining the sequence required for a given function is not straightforward. We are motivated in particular by applications in bioenergy, where engineered microbes can be used to synthesize chemical fuels from carbon dioxide. To inform engineering, we use metabolic models to computationally predict and design the metabolism of the microbes. These models are essentially network flow models, and optimizing metabolic flow amounts to a difficult network design problem. We discuss GDLS (Genetic Design through Local Search), a heuristic we have developed for solving this network design problem. Our heuristic, which is related to the M algorithm for limited search trellis decoding of convolutional codes, results in effective, low-complexity search of the design space. We have applied GDLS to iAF1260, a detailed metabolic model of Escherichia coli, to predict modifications of the E. coli genome for overproduction of fatty acids-an intermediate to hydrocarbon fuels-and we are currently in the process of implementing these modifications and validating the experimental phenotype.