Enzymatic target identification with dynamic states

• Authors: Bin Song;Sanjay Ranka;Tamer Kahveci
• Venue: Proceedings of the First ACM International Conference on Bioinformatics and Computational Biology
• Year: 2010

As a metabolic network reaches from a state to a steady state, a subset of its fluxes gradually change. The sequence of intermediate states, called the dynamic state, shows the pattern that the underlying network follows to reach the steady state Understanding this pattern is crucial for metabolic engineering as the intermediate states can lead to undesired effects such as toxicity although the final steady state is a desired one. In this paper, we consider the problem of enzymatic target identification in metabolic networks. Unlike existing strategies, we consider the dynamic behavior of the state changes of the networks. Given a goal pattern for the fluxes of a given network, we aim to find the set of enzyme knockouts that will produce a dynamic state similar to that goal pattern. We consider three distance functions to measure the difference between two dynamic states. These are the Euclidean distance, time-warping distance and pattern distance. Euclidean distance restricts the solution space to the exact goal state. Time-warping distance allows for stretching of the goal pattern in the time domain. Pattern distance allows scaling and shifting of the goal flux in addition to stretching in the time domain. We provide a branch and bound method to solve this problem. We also develop a partitioning strategy to reduce the running time of our method. This strategy avoids constructing the entire dynamic state by computing a lower bound to the distance between two dynamic states when the entire dynamic state is not available. Our experiments on the purine metabolism show that our method runs accurately. They also show that our partitioning strategy reduces the number of time intervals computed for dynamic states by a factor of 2 to 6.