Knockout Prediction for Reaction Networks with Partial Kinetic Information

International audienceIn synthetic biology, a common application field for computational methods is the prediction of knockout strategies for reaction networks. Thereby, the major challenge is the lack of information on reaction kinetics. In this paper, we propose an approach, based on abstract interpretation, to predict candidates for reaction knockouts, relying only on partial kinetic information. We consider the usual deterministic steady state semantics of reaction networks and a few general properties of reaction kinetics. We introduce a novel abstract domain over pairs of real domain values to compute the differences between steady states that are reached before and after applying some knockout. We show that this abstract domain allows us to predict correct knockout strategy candidates independent of any particular choice of reaction kinetics. Our predictions remain candidates, since our abstract interpretation over-approximates the solution space. We provide an operational semantics for our abstraction in terms of constraint satisfaction problems and illustrate our approach on a realistic network. An extended version of this paper with appendices containing detailed information on proofs and on the example model is provided at http://www.lifl.fr/BioComputing/extendedPapers/vmcai13.pdf. Keywords: Abstract interpretation, deterministic semantics, steady state, constraint satisfaction, synthetic biology