Use this URL to cite or link to this record in EThOS: http://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.653373
Title: Statistical methods for pre-processing microarray gene expression data
Author: Khondoker, Md. Mizanur Rahman
Awarding Body: University of Edinburgh
Current Institution: University of Edinburgh
Date of Award: 2006
Availability of Full Text:
Access from EThOS:
Full text unavailable from EThOS. Please try the link below.
Access from Institution:
Abstract:
A novel method is developed for combining multiple laser scans of microarrays to correct for “signal saturation” and “signal deterioration” effects in the gene expression measurement. A multivariate nonlinear functional regression model with Cauchy distributed errors having additive plus multiplicative scale is proposed as a model for combining multiple scan data. The model has been found to flexibly describe the nonlinear relationship in multiple scan data. The heavy tailed Cauchy distribution with additive plus multiplicative scale provides a basis for objective and robust estimation of gene expression from multiple scan data adjusting for censoring and deterioration bias in the observed intensity. Through combining multiple scans, the model reduces sampling variability in the gene expression estimates. A unified approach for nonparametric location and scale normalisation of log-ratio data is considered. A Generalised Additive Model for Location, Scale and Shape (GAMLSS) is proposed. GAMLSS uses a nonparametric approach for modelling both location and scale of log-ratio data, in contrast to the general tendency of using a parametric transformation, such as arcsinh, for variance stabilisation. Simulation studies demonstrate GAMLSS to be more powerful than the parametric method when a GAMLSS location and scale model, fitted to real data, is assumed correct. GAMLSS has been found to be as powerful as the parametric approach even when the parametric model is appropriate. Finally, we investigate the optimality of different estimation methods for analysing functional regression models. Alternative estimators are available in the literature to deal with the problems of identifiability and consistency. We investigated these estimators in terms of unbiasedness and efficiency for a specific case involving multiple laser scans of microarrays, and found that, in addition to being consistent, named methods are highly efficient and unbiased.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID: uk.bl.ethos.653373  DOI: Not available
Share: