Type

Text

Type

Dissertation

Advisor

Nancy R. Mendell | Finch, Stephen J. | Wei Zhu | Derek Gordon.

Date

2010-08-01

Keywords

Statistics

Department

Department of Applied Mathematics and Statistics

Language

en_US

Source

This work is sponsored by the Stony Brook University Graduate School in compliance with the requirements for completion of degree.

Identifier

http://hdl.handle.net/11401/72659

Publisher

The Graduate School, Stony Brook University: Stony Brook, NY.

Format

application/pdf

Abstract

Variation in the human genome is present in many forms, including single-nucleotide polymorphisms (SNPs) and copy number polymorphisms (CNPs). CNPs have many categories such as small insertion-deletion polymorphisms, variable number of repetitive sequences, and genomic structural alterations. A major question that researchers in the field of statistical genetics need to answer is the number of CNP categories in a given dataset. In this study, I compare five information criteria (BIC, AIC, NEC, CLC, and ICL-BIC) to find if there is a"best" measure among them in finding the correct number of components (correct number of CNP categories). I consider six design factors: equal/unequal within-component variances, high/low separations, sample size, mixture proportion, multiple random starting values, and transformation using two known number of components (3 and 6). The result indicates that under"ideal" conditions (that is, small number of components, large separation between components, constant within component variance, and no subsequent transformation of mixture data), each criterion performs well. When the data is a monotonic transformation of data from a mixture, the BIC criterion, which is the most commonly used criterion in CNP research, has a low component number accuracy rate. I then considered the application of the Box-Cox transformation whether or not it was needed. The application of the Box-Cox transformation did not reduce the component number accuracy rate of the CLC, ICL-BIC, and BIC when it was not needed. The component number accuracy rates for the BIC criterion with Box-Cox transformation applied were improved when the mixture data was transformed. The Box-Cox transformation should be used routinely with CLC, ICL-BIC, or BIC criterion to estimate the number of components in a CNP mixture analysis.

Share

COinS
 
 

To view the content in your browser, please download Adobe Reader or, alternately,
you may Download the file to your hard drive.

NOTE: The latest versions of Adobe Reader do not support viewing PDF files within Firefox on Mac OS and if you are using a modern (Intel) Mac, there is no official plugin for viewing PDF files within the browser window.