Integrative modelling of stochasticity noise heterogeneity and measurement error in the study of model biological systems

Recent breakthroughs in experimental technology have allowed the study of the dynamics of cell biochemistry at single-cell resolution. These studies confirm earlier theoretical predictions that such dynamics would be intrinsically stochastic. Such randomness in cellular behaviour has now been confirmed to be an important component in the observed heterogeneity of genetically identical cells cultured in a uniform environment. However, intrinsic stochasticity is just one source of heterogeneity in biological data, and others, such as minor unavoidable variations in environment and limitations in the measurement technology, can have an equally important effect. A major goal of modern biology is to build dynamic, predictive, quantitative models of the behaviour of biological systems. Computational models enable in silico testing of plausible biological hypotheses and help to establish a clearer understanding of the complex genetic and biochemical mechanisms at play. Despite technological advances, much biological data used to build, refine and test models consists of measurements on cell populations. Unless the models that we build properly reflect the multiple sources of heterogeneity in such data, it is difficult to use them in order to test model adequacy or refine the model parameters or structure to more accurately reflect the underlying biology. The only coherent framework for mathematically modelling noise and heterogeneity is based on probability theory, and especially the theory of stochastic processes. Single-cell stochastic kinetic models are already established as a powerful tool for modelling intrinsic noise in simple genetic and biochemical networks. However, in such cases the emphasis is almost exclusively focused on the effect of noise on single-cell dynamics. It is important to now move beyond this and build integrated stochastic models that incorporate multiple sources of noise and heterogeneity. Stochastic models which integrate multiple levels of organisation are therefore a vital part of the vision for Systems Biology. This proposal comes from a statistician already expert in the necessary mathematical, computational and stochastic modelling techniques. The plan is to train him in modern molecular and cell biology, and associated experimental techniques. The idea is not to 'convert' him into a bench biologist, but by spending time in the lab he will better appreciate the practical issues confronted by 'wet lab' scientists, and will be in a much better position to be able to accurately model the multiple sources of heterogeneity that give rise to experimental data. This will allow the development of a number of examples of integrative stochastic models of biological systems, in order to demonstrate the utility of the approach.

The proposal is to train an expert stochastic modeller (but novice biologist) in modern cell and molecular biology theory and experimental techniques, primarily through embedding in two excellent labs. One lab works on studying noise and heterogeneity in Bacillus subtilis competence development, at the single-cell level, via fluorescence microscopy live cell imaging and flow cytometry, using GFP reporters. As well as shadowing research assistants and technicians, the fellow will conduct a small experimental programme of his own, and use the insights gained for building more realistic models of the bi-stable Bacillus competence system. The models developed will lead to new hypotheses about the mechanisms underlying competence development. The models will also suggest how the hypotheses can be tested in the lab, leading to further refinement of the models. The second lab works on studying the cellular response to telomere uncapping in Saccharomyces cerevisiae. As well as using a variety of more conventional molecular biology techniques, the lab specialises in the use of high-throughput technology for large-scale study of genetic and environmental effects on the damage response phenotype. In particular, they have a state-of-the-art robotic system for genome-wide screening of mutants which generates large amounts of semi-quantitative data with a complex error structure. Accurate statistical modelling of the whole-system behaviour is a non-trivial challenge, and requires a detailed working knowledge of the robot. Again, the fellow will begin by shadowing members of the lab before conducting his own genome-wide screening experiments on the robot. The experience will be used to build accurate statistical models of the data, and then use them to conduct inference for mechanistic stochastic kinetic models of the telomere-uncapping response. Again, an iterative process of modelling and lab investigation of in silico predicted hypotheses will be employed.