Control and optimization methods in biomedical systems: from cells to humans
Optimization and control theory are well developed techniques to quantize, model, understand and optimize real world systems and they have been widely used in engineering, economics, and science. In this thesis, we focus on applications in biomedical systems ranging from cells to microbial communities, and to something as complex as the human body. The first problem we consider is that of medication dosage control for drugs delivered intravenously to the patient. We focus specifically on a blood thinner (called bivalirudin) used in the post cardiac surgery Intensive Care Unit (ICU). We develop two approaches (a model-free and a model-based one) that predict the effect of bivalirudin. After obtaining the model and its best fit parameters by solving a non-linear optimization problem, we develop automatic dosage controllers that adaptively regulate its effect to desired levels. Our algorithms are validated using actual data from a large hospital in the Boston area. In the second problem, we introduce a cellular objective function inference mechanism in metabolic networks. We develop an inverse optimization method, called InvFBA (Inverse Flux Balance Analysis), to infer the objective functions of growing cells by using their reaction fluxes. InvFBA can be seen as an inverse version of FBA (Flux Balance Analysis) which predicts the distribution of the cell's reaction fluxes by using a hypothetical objective function. The objective functions can be linear, quadratic and non-parametric. The efficiency of the InvFBA approach matches the structure of the FBA and ensures scalability to large networks and optimality of the solution. After testing our algorithm on simulated E. coli data and time-dependent S. oneidensis fluxes inferred from gene expression data, we apply our inverse approach to flux measurements in long-term evolved E. coli strains, revealing objective functions that provide insight into metabolic adaptation trajectories. In the final problem in this thesis, we formulate a novel resource allocation problem in microbial ecosystems. We consider a given number of microbial species living symbiotically in a community and a list of all metabolic reactions present in the community, expressed in terms of the metabolite proportions involved in each reaction. We are interested in allocating reactions to organisms so that each organism maintains a minimal level of growth and the community optimizes certain objectives, such as maximizing growth and/or the uptake of specific compounds from the common environment. We leverage tools from Flux Balance Analysis (FBA) and formulate the problem as a mixed integer linear programming problem. We test our method in a toy model involving two organisms that can only survive through cross-feeding, demonstrating that the method can recover this interaction. We also test the method in a community of two simplified bacteria described in terms of their core, simplified metabolic network. We demonstrate that the method can obtain syntrophic cross-feeding species that would be very difficult to design manually.