Towards a Systems Understanding in Biology: Data and Modeling
Breast cancer signaling pathways and neural systems are composed of networks of mutually dependent and thus interconnected genes or neurons. Systems biology is an emerging interdisciplinary field of study that focuses on understanding these complex interactions using a systematic approach. The main goal is transforming biology into a fully quantitative, theory-rich science to understand complex behavior and produce effective predictions. However, since often only incomplete abstracted hypotheses exist to explain observed complex behavior and functions, new techniques for identifying breast cancer signaling pathways and neural systems need to be able to incorporate different types of experimental data and varying levels of prior knowledge.
This dissertation describes several efforts aimed towards an understanding of these systems by developing system identification tools and computational analysis techniques to facilitate these studies. In systems biology, models may focus on different features, different aspects and different objectives, so there can exist several model classes.
The first class of tools that we present includes the hybrid Boolean framework and optimization-based inference, and is developed in conjunction with existing mathematical tools. For example, the former is a combination of ordinary differential equations (ODEs) and Boolean networks (BNs), and the latter is that of graph models and linear time-varying systems. These tools take advantage of existing mathematical models and compensate for their limitations or disadvantages. The second class of tools that we present includes data-driven graph reconstruction using compressive sensing, discrete mode identification via sparse subspace clustering, and low-rank representation of neural activity. These adopt a data-driven approach using little prior knowledge which has its origin in the computer vision literature. Case studies of Human Epidermal growth factor Receptor 2 overexpressed breast cancer and neural systems for a Brain Machine Interface are given.
Since the challenge in systems biology has become to show that the identified networks and corresponding mathematical models are enough to represent the underlying system, these tools are developed both for identifying models and also suggesting experimental directions to better understand the systems. The ultimate goal of the work presented is to create a framework to broaden the spectrum of different modeling approaches since each model can be described through a different perspective and may highlight different aspects of the system.