A differential gene expression algorithm for comparative microarray analysis

The novel Differential Gene Expression (DGE) algorithm for the comparative analysis of microarrays presented here provided a powerful method for the determination of changes in gene expression (both upregulation and downregulation) under control and experimental conditions. The DGE algorithm which used the Nonlinear Fold-Change (NFC) method for assessing changes in gene expression was shown to improve upon two types of algorithms: 1. precedent differential gene expression algorithms which used the Linear Fold-Change (LFC) method and 2. Significance Analysis of Microarrays (SAM) algorithm which used the modified T-statistic method. The NFC method which was based on nonparametric quantile regression analysis was implemented using the kernel weighted local linear fitting algorithm to approximate conditional quantile functions of gene expression. The nonparametric quantile regression analysis was used for the estimation of the gene ranks. The generation of the more stable gene rank estimates led to the more effective identification of differential gene expression. The SAM algorithm and the DGE algorithm were subjected to comparative accuracy performance analysis and comparative precision performance analysis. The DGE algorithm offered five distinct advantages over the SAM algorithm: 1. increased accuracy, 2. increased precision, 3. automated differential criterion selection, 4. increased statistical gene expression measurement categories and 5. more robust statistical gene expression measurement classification error fault tolerance. While the DGE algorithm represented an improvement over the current SAM algorithm, the DGE algorithm nevertheless demonstrated four limitations: 1. terminal quantile computation, 2. terminal expression level measurement computation, 3. local neighborhood computation and 4. time complexity analysis. Five future research directions for extending the DGE algorithm were proposed: 1. exploration of statistical gene expression measurement properties, 2. development of an extended nonparametric quantile regression algorithm, 3. development of an extended local neighborhood algorithm, 4. parallel computer implementation and 5. calculation of receiver operating characteristic curves.