Target practice: informatic and metabolomic assessment of biological network changes and of drug-cell interactions

There are many occasions where one may wish to know the site of interaction of a drug or other substance with a complex biological system (i.e. network), typically by detecting changes something that can be measured. These are usually hard problems, since there are many ways of explaining the changes if one does not in fact know the network. However, in contrast to biological signalling and gene regulatory networks, we normally DO know the structure and outline properties of METABOLIC networks. This makes it MUCH easier to determine the parts of the network changes might have been focussed. Initially using baker's yeast as a model organism, we wish to demonstrate that this strategy does indeed work. It is necessary to make simple assumptions about the general form of the the equations describing the interactions within these networks, and we shall develop and exploit modern numerical methods for parameter estimation. As stated, we shall initially develop and test these strategies in baker's yeast, Saccharomyces cerevisiae, since this is a well understood organism. However, our collaborative partner Unilever are extremely interested in Corynebacterium jeikeium, for which a genome sequence and network model exist, and using resources made available by them for this project we shall also exploit these methods in the analysis of metabolic fluxes in this organism. The deliverable will be a suite of novel methods with which to infer the site of action of any drug-like molecule in a reasonably well understood metabolic network.

There are many occasions where one may wish to know the site of interaction of an effector molecule with a complex biological system (i.e. network), typically by measuring changes in the accessible state variables. These are usually ill-conditioned problems, in the sense that many models can account for the observable data, and to make progress it is necessary to apply constraints and simplifications of various kinds. In contrast to cognate analyses of signalling and gene regulatory networks, the analysis of METABOLIC networks and their fluxes is attractive since they NECESSARILY possess stoichiometric and thermodynamic constraints, which are known, and measurement of the molecules they excrete as end products creates further constraints on the fluxes through the different parts of the network. Initially using baker's yeast as a model organism, we wish to demonstrate that this strategy does indeed work. The necessary simplifications include the use of mass action and lin-log kinetics, while we shall develop and exploit modern methods of multivariate statistical optimisation and machine learning for parameter estimation. These include multi-objective evolutionary algorithms, and the exploitation of probabilistic graphical methods and Gaussian process models. We shall initially develop and test these strategies in baker's yeast, Saccharomyces cerevisiae, since this is a well understood organism. However, our collaborative partner Unilever are extremely interested in Corynebacterium jeikeium, for which a genome sequence and network model exist, and using resources made available by them for this project we shall also exploit these methods in the analysis of metabolic fluxes in this organism. The deliverable will be a suite of novel methods with which to infer the site of action of any effector in a reasonably well understood metabolic network.