Aristotle University Of Thessaloniki (AUTH)
Στατιστικές μέθοδοι για τη μελέτη της μεταβλητότητας και των δομών συσχετίσεων βιοϊατρικών δεδομένων
Abstract
dc:descriptionStatistical science is inextricably linked to the biological sciences, where it is used as a tool to help draw conclusions but also to understand diseases and biological functions in general. Clinical trials and survival models are commonly described as biostatistical methods. For many years, however, a plethora of other statistical methods have been applied to medical and biological data.The general area of research of this dissertation is the application of statistical methods to biological data. More specifically, we tried to use methods that were not widely used in similar data in order to discover homogeneous data groups, to study the correlation structures of different groups, and for prediction. The data used can be grouped into three categories: a) gene expression for genes involved in a disease, b) EEG signals for the human brain's response to a stimulus, and c) sensor measurements, which record the movements of people in order to recognize activities. The statistical methods that were the main object of study of this dissertation are the Archetypal analysis, Structural Equation Models and comparisons of correlations. Archetypal Analysis is a method that has not received the proper attention and groups observations based on their relationship to archetypes. Archetypes are some extreme cases that can be real observations. Structural Equation Models and the comparison of correlation coefficients, study correlation or covariance structures. Structural Equations Models belong to the field of multivariate analysis and have the ability to process many equations simultaneously. In this way they can treat a variable as independent in an equation and as dependent in another equation. Comparison of correlations involves some statistical tests that examine the statistically significant difference between two correlation coefficients. It is a method that is not often used, but it is very useful in the study of heterogeneous correlation structures.Archetypal Analysis has been applied to neurological responses describing the brain activity of individuals exposed to a stimulus, for grouping them into homogeneous groups, and to clinical and biological data of patients who have undergone assisted reproduction therapy to identify which hormones contribute to the success of therapy. Structural Equation Models have been used to detect gene signaling pathways in patients with Chronic Lymphocytic Leukemia. This patient cohort consisted of two groups of patients with different clinical picture. Using Structural Equation Models, most relationships between genes were modeled and differences were identified in the signaling pathways of the two patient groups. The study of correlation structures was used in both gene expression and measurement of wearable sensors to identify activity. Based on the comparison of correlations, we proposed a new method of selecting variables for classification algorithms.The dissertation is conceptually divided into two parts. The first part concerns data grouping and includes Chapters 1 to 3. In these chapters the method of Archetypal Analysis is revised and its two applications are described. The second part deals with the study of correlation and covariance structures, which was the main part of the research and consists of Chapters 4 to 9. Two chapters describe the theory of Structural Equation Models and methods of comparing correlation coefficients and the rest concern the applications of these methods.We could summarize the main points of the dissertation's contribution to the following: a) the discovery of relationships between genes of heterogeneous groups of patients with Chronic Lymphocytic Leukemia, using Structural Equation Models b) the suggestion of an alternative way of grouping brain responses using Archetypal Analysis, c) the proposal of a new method of selecting variables based on the comparison of correlations of different observation groups.
Degree
thesis:*- Grantor dc:publisher
- Aristotle University Of Thessaloniki (AUTH)
- Year dc:date
- 2020
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Tsanousa, Athina
- Τσανούσα, Αθηνά
Subjects
dc:subject × 20- Συσχέτιση
- Μοντέλα δομικών εξισώσεων
- Ομαδοποίηση δεδομένων
- Ομαδοποίηση γράφων
- Επιλογή μεταβλητών
- Μεταβλητότητα
- Βιοστατιστική
- Correlations
- Structural equation modeling
- Clustering
- Graph clustering
- Feature selection methods
- Variability
- Biostatistics
- Μαθηματικά
- Φυσικές Επιστήμες
- Στατιστική και Πιθανότητες
- Mathematics
- Natural Sciences
- Statistics and Probability
Rights
- Language dc:language
- gre
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/49417