The thesis topic is the analysis of commonly used measurement and evaluation methods, error analysis of these methods, and the design of new methods for learning process description and elimination of identified errors. The analyses focuses on motor and spatial learning experiments of inbred mice in association with neurodegenerative diseases research. Inadequacies of standard measurement and evaluation methods as well as usefulness of overlooked methods from biological statistics and classification criteria are presented. New evaluation methods are designed based on approximations using either the normal, Weibull or Cox phase distribution. Together with the evaluation methods, the influence of non-use of a priori information is determined, sensitivity analysis with respect to the size of the data set is performed and errors of the methods and measurement are identified. The primary outcomes are protocols for the evaluation of experiments and learning process descriptions that bare either fundamental or minimal diferences from standard solutions.
The use of animal (biological) models is the important part of biomedical research. In the case of neurodegenerative diseases, there are especially used mice, rat, and nonhuman primates. Tools for determining the level of functions/dysfunctions related to neurodegenerative diseases and using mice models are several methods of motor and spatial learning, related with CNS excitability. The research is then associated with the evaluation of the learning process, motor and cognitive abilities changes concretely. The thesis deals with the approach of the learning process evaluation models design, constructed specifically for using experimental animal models.
The thesis presents the inadequacies of standard measurement and evaluation methods and proposes possible innovations in gathering information for comprehensive evaluations of experiments and learning processes (e.g., non-standard use of correlation analysis, analysis of variance and quantile information, etc.).
The system representation of the learning process is described in relation with the performed error analysis, necessary for the evaluation of the tested hypotheses using common methods of hypotheses testing. A classification criterion is proposed for proper decomposition of the dataset. For description of a new approach of experiments evaluation is defined the investigated system in terms of stochastic systems. The learning process obtained from mice cognitive and motor function abilities measurement is described by random variables (for attributes success rate and time transition, generally latency). The fundamental change to the new approach compared to that commonly used is using both learning system attributes values (a success rate and a time transition) for each motor or spatial learning test evaluation with no use of normal distribution parameters but with using empirical distribution function and its approximation by suitable theoretical distribution. Approximation methods of empirical distrbutions are designed, a criterion for the selection of appropriate approximation is defined and deviations of used approximations are identified.
Novelty of designed approach lies in the evaluation of the learning process in probability and physical units of random variables by using Wasserstein pseudometrics (via determination of the statistical distance). Proposed are also means of better utilizing informational innovations of individual experimental tests, yield and learning quality. Effects of a priori information omissions are determined, sensitivity analysis with respect to the size of the data set is performed and errors of methods and measurements are identified.
The primary outcomes of the thesis are protocols for the evaluation of experiments and learning process descriptions that bare either fundamental or minimal differences from standard solutions. The proposed methods were validated on several experiments performed by Faculty of Medicine in Pilsen.