Multivariate analysis as a tool for understanding and reducing complexity of mathematical models in systems biology
Year: 2011
Type: dissertation
Abstract: In the area of systems biology, technologies develop very fast, which allows us to collect massive amounts of various data. The main interest of scientists is to receive an insight into the obtained data sets and discover their inherent properties. Since the data often are rather complex and intimidating equations may be required for modelling, data analysis can be quite challenging for the majority of bio-scientists who do not master advanced mathematics. In this thesis it is proposed to use multivariate statistical methods as a tool for understanding the properties of complex models used for describing biological systems.\nThe methods of multivariate analysis employed in this thesis search for latent variables that form a basis of all processes in a system. This often reduces dimensions of the system and makes it easier to get the whole picture of what is going on. Thus, in this work, methods of multivariate analysis were used with a descriptive purpose in Papers I and IV to discover effects of input variables on a response.\nOften it is necessary to know a functional form that could have generated the collected data in order to study the behaviour of the system when one or another parameter is tuned. For this purpose, we propose the Direct Look-Up (DLU) approach that is claimed here to be a worthy alternative to the already existing fitting methods due to its high computational speed and ability to avoid many problems such as subjectivity, choice of initial values, local optima and so on (Papers II and III).\nAnother aspect covered in this thesis is an interpretation of function parameters by the custom human language with the use of multivariate analysis. This would enable mathematicians and bio-scientists to understand each other when describing the same object. It was accomplished here by using the concept of a metamodel and sensory analysis in Paper IV. In Paper I, a similar approach was used even though the main focus of the paper was slightly different. The original aim of the article was to show the advantages of the multi-way GEMANOVA analysis over the traditional ANOVA analysis for certain types of data. However, in addition, the relationship between human profiling of data samples and function parameters was discovered.\nIn situations when funds for conducting experiments are limited and it is unrealizable to study all possible parameter combinations, it is necessary to have a smart way of choosing a few but most representative conditions for a particular system. In Paper V Multi-level Binary Replacement design (MBR) was developed as such, which can also be used for searching for a relevant parameter range. This new design method was applied here in Papers II and IV for selection of samples for further analyses.
Source: NVA (Nasjonalt vitenarkiv)
Author: Julia Isaeva
Institution: Norwegian University of Life Sciences
Language: English
Subfield: Statistics and Probability
Field: Mathematics
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W2617699342
