|dc.contributor.author||Naiwala Pathirannehelage, Kaushala Samudini Jayawardana||-|
|dc.description.abstract||Statistics in medical research gained a vast surge with the development of high-throughput biotechnologies that provide thousands of measurements for each patient. These multi-layered data has the clear potential to improve the disease prognosis. Data integration is increasingly becoming essential in this context, to address problems such as increasing the power, inconsistencies between studies, obtaining more reliable biomarkers and gaining a broader understanding of the disease. This thesis focuses on addressing the challenges in the development of statistical methods while contributing to the methodological advancements in this field.
We propose a clinical data analysis framework to obtain a model with good prediction accuracy addressing missing data and model instability. A detailed pre-processing pipeline is proposed for miRNA data that removes unwanted noise and offers improved concordance with qRT-PCR data. Platform specific models are developed to uncover biomarkers using mRNA, protein and miRNA data, to identify the source with the most important prognostic information.
This thesis explores two types of data integration: horizontal; the integration of same type of data, and vertical; the integration of data from different platforms for the same patient. We use multiple miRNA datasets to develop a meta-analysis framework addressing the challenges in horizontal data integration using a multi-step validation protocol. In the vertical data integration, we extend the pre-validation principle and derive platform dependent weights to utilise the weighted Lasso. Our study revealed that integration of multi-layered data is instrumental in improving the prediction accuracy and in obtaining more biologically relevant biomarkers. A novel visualisation technique to look at prediction accuracy at patient level revealed vital findings with translational impact in personalised medicine.||en_AU|
|dc.publisher||University of Sydney||en_AU|
|dc.publisher||Faculty of Science||en_AU|
|dc.publisher||School of Mathematics and Statistics||en_AU|
|dc.title||Prognostic Methods for Integrating Data from Complex Diseases||en_AU|
|dc.type.pubtype||Doctor of Philosophy Ph.D.||en_AU|
|Appears in Collections:||Sydney Digital Theses (Open Access)|