Developing methods for inferring regulatory mechanisms from intact systems: a neisseria case study

The behaviour of biological systems is controlled and coordinated through a network of 'regulators' and other intracellular interactions that control the expression of the genes within the cell. In bacteria the production of the messenger RNA and the production of proteins are closely linked, and much of the way in which a cell's behaviour is controlled is done at the level of transcription. Transcription can be measured for all of the genes in a cell simultaneously, using microarrays, and this gives a relatively direct read-out of the way in which many aspects of the cell's behaviour are being controlled. This method gives a 'snap shot' of the transcribed genes, and when the observations from many snap shots are combined, the way in which the cell controls its functions can be progressively pieced together, much as the meaning of a movie can be pieced together from the combination of multiple 'frames'. Finding ways to use this information to make testable models that can be used to dissect these central processes that control biological systems is a critical component of systems biology and understanding how biological systems work at a fundamental and 'whole cell' level. If causal relationship and key interactions controlling a cell's behaviour can be determined based upon this type of 'observational' information, then this means that these systems can be addressed without the (frequently impossible, impractical, or unaffordable) need to address each gene individually. Many genes are required for a cell to survive. Other genes are not required for life, but the resulting cell does not function 'normally' in several ways when a normal component has been removed / and it is very difficult to tell which effects are directly or indirectly due to the effects of a gene / gene product. To understand how cellular systems work, we propose that we need ways to analyze and use the information from 'intact / unbroken' biological systems. In this proposal we will make use of one of the largest collections of 'transcript' data, and augment this with information specifically designed to assist modeling the ways in which the cell is controlled. The effectiveness of this modeling will be tested, and the models will be augmented and refined by addressing the key genes by making mutants and testing to what extent they behave according to the model predictions. In this way, we will develop a generally applicable approach that can be applied generally, without the need for expensive, time consuming, and potentially misleading mutant generation in the future.

We have a unique collection of existing microarray data from which to address the regulatory networks of a bacterial system. This collection of data, already with more than 300 channels of high quality, semi-quantitative data, non-dye-incorporation biased data, which we have validated as both dual-channel and single-channel data, will form the basis for initial modeling. In addition, a model of the interactions between about 30% of the main neisserial regulators has already been developed, from the analysis of classical direct comparisons of mutant and wild-type strains, which will be used to test model predictions and form a basis for building a model of regulatory networks. The mathematical analysis will involved two complementary approaches which are applicable to de novo and hypothesis/model driven analysis or transcriptional regulation, respectively. We will use graphical models, building on existing work in graphical Gaussian models and (dynamical) Bayesian Network models to infer which interactions are likely to occur. This work will be supplemented by analysis of mechanistic models. Both approaches will yield testable predictions for regulatory relationships. Model predictions will be tested addressing key genes at 'causal nodes' and those with reciprocal relationships, suggestive of negative feedback, using deletion mutants. These will be generated, using context appropriate cassettes, and they will be expression profiled, using the same high-quality data generating methods, to both test and validate the modeling approaches, and to extend the datasets used for the predictive modeling. Ultimately, the modeling will be built upon over 1000 channels of expression data, and will provide a real and challenging test of this experimental and analytical approach to the analysis of biological systems.