Off-campus WSU users: To download campus access dissertations, please use the following link to log into our proxy server with your WSU access ID and password, then click the "Off-campus Download" button below.
Non-WSU users: Please talk to your librarian about requesting this dissertation through interlibrary loan.
Date of Award
There are many needs to populate bioinformatics databases for its professional purposes to build a valuable view of various genomic databases over the Internet. One of these studies is to testify a way to build an organism-specific protein interaction database. The multilayered architecture of the designed system has wrapper, mediator, relational database, and user interface components. Even though neural network has been proven a good approach in machine learning, it has not been popular in proteomic studies because of categorical data description and data size in those applications. In order to avoid these problems, this study adopts decimal-to-binary coding scheme and class decomposition in data analysis. Each decomposed subproblem constructs a modular neural network to predict functions of proteins without known function. The experiments show that this modular neural network approach benefits of fast learning and high accuracy. Another study for predictive model of protein function is based on instance-based learning. The studied method fully uses the concept of "guilt-by-interaction" in assigning functions to proteins without known function. The mechanism to choose reliable interactions is embedded in the proposed method. In experiments, the accuracy of a modular neural network can be improved through an ensemble methodology with an instance-based learning.
Hwang, Dooseoung., "Computational meta-analysis of protein interaction data" (2003). Wayne State University Dissertations. 3402.