Seminar in Biology for CSE

Abstract

The seminar explores a range of computational and modeling techniques from dynamic systems, control theory, machine learning, and pharmacometrics in the context of type 1 diabetes. The students conducts a literature study on a relevant topic chosen in agreement with the supervisor, critically appraises the literature/methods in a brief written report and presents findings in an oral presentation.

Objective

Students will learn to evaluate and present scientific literature and critically appraise proposed methods and models. The focus on a single application area (type 1 diabetes) allows comparison of methods from different computational fields and provides awareness of strengths, weaknesses, but also commonalities of different methodological approaches.

Content

Type 1 diabetes is a chronic disease caused by autoimmune destruction of insulin-producing beta cells. Patients require exogenous administration of insulin to provide glucose homeostasis, and insulin treatment regimes need to cover both basal insulin needs, handle large disturbances to glucose homeostasis caused by meals, and be robust to additional disturbances from, e.g. exercise or diurnal rhythms. Mathematical modeling plays a key role in developing insulin therapies and early models range back to the 1980s. Models are used to determine key clinical parameters for individual patients, for predicting glucose dynamics after meals, for adapting insulin dosing and timing, for describing effects of exercise, for controlling automated insulin delivery systems, and for many other purposes. Models and computational techniques include small compartment models typical for pharmacometrics, large dynamic models (usually ordinary differential equations), methods from optimal control and model-predictive control, techniques for signal processing, systems biology models at the cellular level, physiologically-based models of whole organisms, as well as methods from machine learning such as Gaussian processes and reinforcement learning. In contrast to most engineering systems, very little is known about the actual dynamics resulting from physiology which leads to large uncertainties about model structure. In addition, many states and parameters cannot be observed directly, clinical measurements are notoriously noisy, and models and techniques need to cope with large variation between individual patients. Hence, modeling and computation in type 1 diabetes faces formidable practical and methodological challenges. In this seminar, we will consider a selection of modeling and computational techniques for different questions surrounding type 1 diabetes therapy. We will discuss and compare different approaches, see how clinical and scientific questions can guide the choice of techniques and models, and discuss challenges in real-world applications.

Resources