AbstractBiological systems typically generate complex data that encapsulate the dynamics of interactions among measurables over time. To support the formation of insights into time series data from a biological system, there is a requirement to develop new methods that can analyse and translate such complex data into a form that allows trends, patterns, and predictions to be easily viewed, verified and tested. Here, a suite of novel analytical and matrix-based techniques for dynamical systems modelling are developed that are time-efficient and data-driven. These techniques facilitate a range of scientific analyses through novel matrix-based system identification and parameter estimation methods. The inference techniques are fast, optimised, and do not require a priori information to successfully infer network of interactions or automatically construct data-consistent models from data. Two distinct principal (Jacobian and power-law) models (solutions) that are data-consistent may be constructed from a single time series data set. A recast technique has also been developed to reconstruct either one of the principal models from the other, providing support for model interoperability and multiple model integration.
The thesis demonstrates the effectiveness of a new theoretical framework developed to incorporate a modelling and visualization pipeline able to deal with a wide range of time-series data sets relating to complex biological systems. The integrated framework is able to infer and depict interaction networks implicit in time series data in just a matter of seconds and then display the evolution of that network dynamics in response to network perturbation such as drug treatments. Beyond this, there is a broader contribution to the field of biochemical system theory (BST), evidenced by establishing methods for transforming a constructed jacobian model to equivalent power-law models, and vice versa. The effectiveness of these new techniques is demonstrated using artificial time series data samples, simulated pseudo-data of biologically plausible models of real biological systems, and real experimental data derived from biological experiments.
|Date of Award||Jul 2013|
|Supervisor||James Bown (Supervisor)|