Search

This study explores the handling and analyzing of big data in the field of bioinformatics. The focus has been on improving the analysis of public domain data for Affymetrix GeneChips which are a widely used technology for measuring gene expression. Methods to determine the bias in gene expression due to G-stacks associated with runs of guanine in probes have been explored via the use of a grid and various types of cloud computing. An attempt has been made to find the best way of storing and analyzing big data used in bioinformatics. A grid and various types of cloud computing have been employed. The experience gained in using a grid and different clouds has been reported. In the case of Windows Azure, a public cloud has been employed in a new way to demonstrate the use of the R statistical language for research in bioinformatics. This work has studied the G-stack bias in a broad range of GeneChip data from public repositories. A wide scale survey has been carried out to determine the extent of the Gstack bias in four different chips across three different species. The study commenced with the human GeneChip HG U133A. A second human GeneChip HG U133 Plus2 was then examined, followed by a plant chip, Arabidopsis thaliana, and then a bacterium chip, Pseudomonas aeruginosa. Comparisons have also been made between the use of widely recognised algorithms RMA and PLIER for the normalization stage of extracting gene expression from GeneChip data.

The Eora, GTAP and WIOD multiregional input-output (MRIO) databases calculate different national level CO2 consumption-based accounts (CBA). If these outcomes are to be used as evidence in climate policy, analysts need to be confident as to the accuracy of the databases and to understand why the results differ. This thesis explores the different data sources, database structures and construction techniques used to build Eora, GTAP and WIOD. Analytical techniques, such as matrix difference statistics, structural decomposition analysis and structural path decomposition are used to quantify the nature of the difference and determine the cause of outcome difference. To make meaningful comparisons between the three MRIO databases, each is mapped to a consistent classification system comprising 40 countries and 17 sectors. The effect of this aggregation is shown to be fairly minimal, giving confidence that the aggregated versions of each database reflect the full-sized versions. This study finds that the main cause of difference in the CO2 CBA as calculated by different MRIO databases lies in the different emissions extension vectors used. Not only is the global emissions total different, but the distribution of emissions by industrial and the household sector differs depending on whether the particular database takes the territorial or residence principle to emissions allocation. The effect of differing global totals can be observed in the national CO2 CBA calculated for the same country being different in each database. The effect of the territorial or residence principle is evident when results are compared at the supply chain level. At this level of detail, it is also possible to quantify the effect of differing construction techniques used to populate data in the economic matrices. The thesis concludes by making recommendations as to how future MRIO databases could be constructed in an accurate and consistent manner and how they should be used in policy in light of the findings.