More and more Bayesian statistical methods are currently employed for the design of knowledge-based systems in areas such as molecular biology, pharmacology and medicine. In particular, Bayesian belief network have gained significant popularity in the field of biomedical informatics.
This thesis centers on the development of methodologies and tools for computerizing disease profiles using generalized Bayesian belief network. Generalized belief networks, which are generalization of classical belief networks, rely on a constraint-based representation of knowledge with the goal to handle knowledge with various degrees of reliability, ranging from very reliable knowledge, guess-style knowledge to fuzzy and uncertain knowledge.
One main objective of this thesis is to implement a software develop tool in helping to generate a generalized Bayesian belief network. An abstract layer is constructed between the user application interface and the belief inference kernel engine. The layer consists of an object-oriented design model and a C++ application interface. A cluster of object classes is implemented to accommodate the definition of topographic structure of a generalized belief network, expression of knowledge uncertainty, specification of knowledge constraints, and the execution of belief inference. Generalized belief networks are implemented using classical belief networks relying on a constraint satisfaction approach. In addition, the thesis reports on case studies that employed generalized belief networks for the generation of disease profiles for Alzheimer disease and Huntington disease.