Computational Methods for Modelling and Analysing Biological Networks

The main theme of this thesis is modelling and analysis of biological networks. Measurement data from biological systems is being produced at such a pace that it is impossible to make use of it without computational models and inference algorithms. The methods and models presented here aim at allowing to extract relevant relationships from the masses of data and formulating complex biological hypotheses that can be studied via simulation. The problem of learning the structure of a popular method class, Bayesian networks, from measurement data is investigated in this thesis, and an improvement to the standard method is presented that facilitates finding the correct network structure. Furthermore, this thesis studies active learning, where the structure inference algorithm can itself suggest measurements to be made. Active learning is applied to realistic scenarios with measured datasets and an active learning method that can deal with heterogeneous data types is presented. Another focus of this thesis is on analysing networks whose structure is known. The utility of a standard method for selecting beneficial mutations in metabolic networks is evaluated in the context of engineering the network to produce a desired substance at a higher rate than normally. Metabolic network modelling is also used in conjunction with a simulation of a biochemical network controlling bacterial movement in a state-based and executable framework that can integrate different submodels. This combined model is then used to simulate the behaviour of a population of bacteria. In summary, this thesis presents improvements on methods for learning network structures, evaluates the utility of an analysis method for identifying suitable mutations for producing a substance of interest, and introduces a state-based modelling framework capable of integrating several submodels.