A statistically robust and biologically-based approach for analysis of microarray data is described that integrates independent biological knowledge and data with a global F-test for finding genes of interest that minimizes the need for replicates when used for hypothesis generation. First, each microarray is normalized to its noise level around zero. The microarray dataset is then globally adjusted by robust linear regression. Second, genes of interest that capture significant responses to experimental conditions are selected by finding those that express significantly higher variance than those expressing only technical variability. Clustering expression data and identifying expression-independent properties of genes of interest including upstream transcriptional regulatory elements (TREs), ontologies and networks or pathways organizes the data into a biologically meaningful system. We demonstrate that when the number of genes of interest is inconveniently large, identifying a subset of "beacon genes" representing the largest changes will identify pathways or networks altered by biological manipulation. The entire dataset is then used to complete the picture outlined by the "beacon genes." This allow construction of a structured model of a system that can generate biologically testable hypotheses. We illustrate this approach by comparing cells cultured on plastic or an extracellular matrix which organizes a dataset of over 2,000 genes of interest from a genome wide scan of transcription. The resulting model was confirmed by comparing the predicted pattern of TREs with experimental determination of active transcription factors.

BACKGROUND: To better understand the response of urinary epithelial (urothelial) cells to Enterococcus faecalis, a uropathogen that exhibits resistance to multiple antibiotics, a genome-wide scan of gene expression was obtained as a time series from urothelial cells growing as a layered 3-dimensional culture similar to normal urothelium. We herein describe a novel means of analysis that is based on deconvolution of gene variability into technical and biological components. RESULTS: Analysis of the expression of 21,521 genes from 30 minutes to 10 hours post infection, showed 9553 genes were expressed 3 standard deviations (SD) above the system zero-point noise in at least 1 time point. The asymmetric distribution of relative variances of the expressed genes was deconvoluted into technical variation (with a 6.5% relative SD) and biological variation components (>3 SD above the mode technical variability). These 1409 hypervariable (HV) genes encapsulated the effect of infection on gene expression. Pathway analysis of the HV genes revealed an orchestrated response to infection in which early events included initiation of immune response, cytoskeletal rearrangement and cell signaling followed at the end by apoptosis and shutting down cell metabolism. The number of poorly annotated genes in the earliest time points suggests heretofore unknown processes likely also are involved. CONCLUSION: Enterococcus infection produced an orchestrated response by the host cells involving several pathways and transcription factors that potentially drive these pathways. The early time points potentially identify novel targets for enhancing the host response. These approaches combine rigorous statistical principles with a biological context and are readily applied by biologists.

SUMMARY: We introduce a novel Matlab toolbox for microarray data analysis. This toolbox uses normalization based upon a normally distributed background and differential gene expression based on five statistical measures. The objects in this toolbox are open source and can be implemented to suit your application. AVAILABILITY: MDAT v1.0 is a Matlab toolbox and requires Matlab to run. MDAT is freely available at http://microarray.omrf.org/publications/2004/knowlton/MDAT.zip.

BACKGROUND: Several aspects of microarray data analysis are dependent on identification of genes expressed at or near the limits of detection. For example, regression-based normalization methods rely on the premise that most genes in compared samples are expressed at similar levels and therefore require accurate identification of nonexpressed genes (additive noise) so that they can be excluded from the normalization procedure. Moreover, key regulatory genes can maintain stringent control of a given response at low expression levels. If arbitrary cutoffs are used for distinguishing expressed from nonexpressed genes, some of these key regulatory genes may be unnecessarily excluded from the analysis. Unfortunately, no accurate method for differentiating additive noise from genes expressed at low levels is currently available. RESULTS: We developed a multistep procedure for analysis of mRNA expression data that robustly identifies the additive noise in a microarray experiment. This analysis is predicated on the fact that additive noise signals can be accurately identified by both distribution and statistical analysis. CONCLUSIONS: Identification of additive noise in this manner allows exclusion of noncorrelated weak signals from regression-based normalization of compared profiles thus maximizing the accuracy of these methods. Moreover, genes expressed at very low levels can be clearly identified due to the fact that their expression distribution is stable and distinguishable from the random pattern of additive noise.

MOTIVATION: We face the absence of optimized standards to guide normalization, comparative analysis, and interpretation of data sets. One aspect of this is that current methods of statistical analysis do not adequately utilize the information inherent in the large data sets generated in a microarray experiment and require a tradeoff between detection sensitivity and specificity. RESULTS: We present a multistep procedure for analysis of mRNA expression data obtained from cDNA array methods. To identify and classify differentially expressed genes, results from standard paired t-test of normalized data are compared with those from a novel method, denoted an associative analysis. This method associates experimental gene expressions presented as residuals in regression analysis against control averaged expressions to a common standard-the family of similarly computed residuals for low variability genes derived from control experiments. By associating changes in expression of a given gene to a large family of equally expressed genes of the control group, this method utilizes the large data sets inherent in microarray experiments to increase both specificity and sensitivity. The overall procedure is illustrated by tabulation of genes whose expression differs significantly between Snell dwarf mice (dw/dw) and their phenotypically normal littermates (dw/+, +/+). Of the 2,352 genes examined only 450-500 were expressed above the background levels observed in nonexpressed genes and of these 120 were established as differentially expressed in dwarf mice at a significance level that excludes appearance of false positive determinations.