| Summary: | Biochemical reaction networks involve many chemical species and are inherently stochastic and complex in nature. Reliable and organised functioning of such systems in varied environments requires that their behaviour is robust with respect to certain parameters while sensitive to other variations, and that they exhibit specific responses to various stimuli. There is a continuous need for improved models and methodologies to unravel the complex behaviour of the dynamics of such systems. In this thesis, we apply ideas from information theory to develop novel methods to study properties of biochemical networks. In the first part of the thesis, a framework for the study of parametric sensitivity in stochastic models of biochemical networks using entropies and mutual information is developed. The concept of noise entropy is introduced and its interplay with parametric sensitivity is studied as the system becomes more stochastic. Using the methodology for gene expression models, it is shown that noise can change the sensitivities of the system at var- ious orders of parameter interaction. An approximate and computationally more efficient way of calculating the sensitivities is also developed using unscented transform. Finally, the methodology is applied to a circadian clock model, illustrating the applicability of the approach to more complex systems. In the second part of the thesis, a novel method for specificity quantification in a receptor-ligand binding system is proposed in terms of mutual information estimates be- tween appropriate stimulus and system response. The maximum specificity of 2 × 2 affinity matrices in a parametric setup is theoretically studied. Parameter optimisation methodology and specificity upper bounds are presented for maximum specificity estimates of a given affinity matrix. The quantification framework is then applied to experimental data from T-Cell signalling. Finally, generalisation of the scheme for stochastic systems is discussed.
|
|---|
