| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
From the DNA Array Unit,
*
Research Resources Branch, National Institute on Aging, National Institutes of Health, Baltimore, Maryland; the Cellular Neurobiology Branch,
National Institute on Drug Abuse, Baltimore, Maryland; and the Department of Psychiatry and Human Behavior,
University of California, College of Medicine, Irvine, California
 |
Abstract |
|---|
 Top Abstract IntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
|
Introduction |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
Before comparing the microarray results from multiple experiments the results from individual experiments must somehow be normalized with respect to each other to account for experimental variation in RNA amounts, specific activity of cDNA labels, and standard handling errors. Failure to properly normalize data used in microarray comparisons runs a high risk of skewing comparison results and reduces the credibility of individual gene change measurements. One of the most common ways in which microarray data are normalized is to assume that the majority of gene expression is relatively constant between experiments and that this constant population can serve as the basis for a general approach to normalization. Empirical observation, in almost all cases, continues to support this underlying assumption used in population normalizations. Occasionally the use of population normalization may be contraindicated when, for example, a highly restricted subset of genes is used to measure a highly dynamic biological condition (eg, a small focused array used to study embryonic development). In this case an alternative normalization method such as spiking of internal references3 should be considered. Clearly, the experimental design must be carefully evaluated before the selection of an appropriate normalization technique.
One basic method of population normalization is global normalization,4 which calculates the mean or median of the signal intensities of each individual experimental dataset and then calculates the mean of the means (or grand mean) for all of the included experiments. Each individual data set is then mathematically adjusted such that the mean of that dataset equals the calculated grand mean. This method is conceptually simple, but when working with datasets having large differences in signal intensity, the data can be inordinately influenced by the presence of outlier data distortions. In addition, and equally as problematic from a high throughput standpoint, each time another experiment is added to the experimental comparisons, the collective or grand mean must be recalculated, and all of the experimental datasets readjusted.
Here we describe the normalization and standardization of cDNA microarray intensity values within datasets by Z score transformation and the subsequent use of the transformed data to compare multiple experiments. The Z score transformation procedure for normalizing data is a familiar statistical method in both neuroimaging5 and psychological studies,6, 7 among others. Recently, Z score transformation statistics have been used in comparing experimental and control group gene expression8, 9, 10 differences by microarray. Z score transformation methods have also been incorporated into the latest version of the public access MAExplorer (supplied by Peter Lemkin of the National Cancer Institute) microarray bioinformatics tool.11
The Z score transformation approach for microarrays corrects data internally within a single hybridization and hybridization values for individual genes are expressed as a unit of SD from the normalized mean of zero. Correction is done before sample-to-sample comparison, and is therefore comparison-independent. Comparisons across samples or across experiments are then performed on equivalently transformed data, and changes in gene expression are expressed as differences between Z scores (Z ratios) or by using a statistical test such as the two-sample-for-means Z test.12 Using this approach, gene expression data derived from different microarray studies becomes comparable across experiments and across laboratories.
In this paper we have compared differences in gene expression using the traditional fold-ratios (arrived at by global normalization) and Z ratios (calculated from Z scores). In addition, we compared significance levels derived from several different statistical methods including Z ratios, Z tests, and t-tests, combined with permutation analysis using Significance Analysis of Microarrays (SAM) software from Stanford University labs13 . We have chosen a dataset that is relatively simple to reduce the number of parameters to be considered in making the comparisons between global and Z score transformations, as well as the comparisons between statistical tests. The Z score transformation process is easily extended to more complex datasets.
|
Materials and Methods |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
RNA Purification
Total cellular RNA was extracted after 2 hours of stimulation directly from T cells in three conditions: control T cells (Ct), T cells following stimulation by phorbol myristate plus ionomycin (PMA+I), and T cells following phorbol myristate plus anti-CD28 antibody (PMA + 28). The total RNA was extracted in the flasks using a one-step guanidine thiocyanate/phenol method14
followed by sequential ethanol precipitations. The concentration and quality of the RNA were assessed by spectrophotometry and by agarose gel electrophoreses. RNA samples were stored at -80°C until used.
RNA Labeling
RNA samples were radiolabeled and hybridized according to protocols described in http://www.grc.nia.nih.gov/branches/rrb/dna.htm. For probe preparation (radiolabeling of total RNA with [33P]dCTP), 5 µg of total RNA for each sample was radiolabeled in a reverse-transcription (RT) reaction. RNA was annealed, in 16 µl H2O, with 1 µg of 24-mer poly(dT) primer (Research Genetics, Huntsville, AL), by heating at 65°C for 10 minutes and cooling on ice for 2 minutes. The RT reaction was performed by adding 8 µl of 5X first-strand RT buffer (Life Technologies, Rockville, MD), 4 µl of 20 mmol/L dNTPs minus dCTP) (Pharmacia, Piscataway, NJ), 4 µl of 0.1 mol/L DTT, 40 U of RNAseOUT (Life Technologies), 6 µl of 3000 Ci/mmol 
[33P]dCTP (ICN Biomedicals, Costa Mesa, CA) to the RNA/primer mixture to a final volume of 40 µl. Two µl (400 U) of Superscript II reverse transcriptase (Life Technologies) was then added, and the sample was incubated for 30 minutes at 42°C followed by additional 2 µl of Superscript II reverse transcriptase and another 30 minutes of incubation. The reaction was stopped by the addition of 5 µl of 0.5 mol/L EDTA. The samples were incubated at 65°C for 30 minutes after addition of 10 µl of 0.1 mol/L NaOH to hydrolyze and remove RNA. The samples were pH-neutralized by the addition of 45 µl of 0.5 mol/L Tris (pH 8.0) and purified using Bio-Rad 6 purification columns (Bio-Rad, Hercules, CA).
Microarray Construction and Use
Microarray construction and hybridization were performed as previously described.15
Briefly, NIA-Immunoarrays, which consist of 1132 genes printed on Nytran + Supercharge nylon membranes (Schleicher & Schuell) in duplicate, were hybridized with [33P]dCTP-labeled cDNA probes overnight at 50°C in 4 ml of hybridization solution. Hybridized arrays were rinsed in 50 ml of 2X SSC and 1% SDS twice at 55°C followed by 1–2 times of washing in 2X SSC and 0.1% SDS at 55°C for 15 minutes each. The microarrays were exposed to phosphorimager screens for 1 to 3 days. The screens were then scanned in a Molecular Dynamics STORM PhosphorImager (Molecular Dynamics, Sunnyvale, CA) at 50 µm resolution. ImageQuant software (Molecular Dynamics) was used to convert the hybridization signals on the image into raw intensity values, and the data thus generated was transferred into Microsoft Excel spreadsheets, pre-designed to associate the ImageQuant data format to the correct gene identities.
Global Normalization
Raw intensity data for each experiment was normalized by first calculating the average intensity for each individual dataset, and then calculating the average of the averages. This grand average was used as the basis for the computation of normalization factors that were subsequently applied to each experiment. The average of all normalized data thereafter equaled the grand average.
Z Score Transformation
Raw intensity data for each experiment is log10 transformed and then used for the calculation of Z scores. Z scores are calculated by subtracting the overall average gene intensity (within a single experiment) from the raw intensity data for each gene, and dividing that result by the SD of all of the measured intensities, according to the formula:
 |
Estimate of Significant Changes in Gene Expression
Traditional Ratio
Traditional ratio calculations of significant changes in gene expression derived from globally normalized data are performed by simply computing the ratio of the average of all of the measurements from one condition or sample to another.4
Significance is customarily assigned to genes whose ratio is greater or equal to 2.0 or less than or equal to 0.5.
Z Ratios
Z score values are used as the data basis in all calculations of changes in gene expression including Z ratios, Z tests, and SAM analysis. Z ratios are calculated by taking the difference between the averages of the observed gene Z scores and dividing by the SD of all of the differences for that particular comparison:
 |
Z Test
An alternative method for calculating significant changes in gene expression, which maximizes the power of replicates and takes into account variation between replicates on a gene by gene basis, is the two-sample-for-means Z test.12
The formula for this statistical test is as follows:
 |
2 is the SD of repeated hybridization intensity measurements (expressed as Z scores) for either condition 1 or condition 2, and n equals the number of repeated measurements for either condition 1 or condition 2. P values can be assigned to the calculated Z test value by consulting the critical Z value for a two-tailed test in a standard normal distribution table.
SAM
SAM (Significance Analysis of Microarrays; software from Stanford University labs13
) analysis was performed on Z score data for two class-unpaired data using the default settings. The samples chosen for analysis came from Donor 1 and included three labeling replicates for control RNA, and two labeling replicates each for the PMA+I and PMA + 28 samples. The SAM procedure combines the calculation of a t-test statistic value for each gene with subsequent permutation analysis and the calculation of a false discovery rate (FDR). Significant gene changes were arbitrarily selected at SAM (d) score values greater than or equal to ± 1.46 (this value yielded the best balance between absolute number of significant calls and the lowest predicted false discovery rate (FDR) for the dataset tested).
Cluster Analysis
Hierarchical clustering of experimental variation in gene expression was determined using software programs developed at Stanford University.16
The cluster algorithm was set to complete linkage clustering using the uncentered Pearson correlation.
|
Results |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
Figure 1
shows a typical hybridization result from these experiments using NIA-Immunoarray filters. Dramatic increases in gene expression between control and stimulated T-cells are illustrated by, for example, the obvious increase in the hybridization signal for interferon-
(IFNgamma). While it is clear from visual inspection that both forms of T-cell activation result in an up-regulation of IFNg, it would also appear from the comparative signal strengths that PMA+I exerts a stronger effect on IFNg gene expression than does PMA + 28. However, it is not possible to reach this conclusion quantitatively without performing some form of normalization procedure before comparing the intensity values derived from the two hybridizations.
|
demonstrates the process for deriving normalized data values using either mean-adjusted (global) normalization or Z score transformation. The data shown here are a subset of the total number of 1132 genes on the NIA Immunoarray, selected for genes that exhibited the most significant increases or decreases between experimental and control samples (the data are sorted in descending order on the PI-Ct column under the Z ratios heading). The first three columns following the gene names contain the raw intensity data before any adjustment and provide the basis for both types of normalization procedures.
|
. Z ratio calculations were also determined for the same genes following Z score transformation. There is a close but not exact correspondence in rank order between the most highly up- or down-regulated genes calculated by either method (see, for example, VIM, P28 vs Ct). Figure 2
illustrates the relationship between the complete datasets of measured Z ratios and fold changes for this condition in the form of a lognormal scatter plot. Linear regression analysis shows that the data fit to a straight line with an r2 value of 0.96. Thus, for this example, estimates of changes in gene expression by the two methods are virtually identical using two distinct normalization procedures (global or Z score transformation) as well as two distinct significance estimates (fold change or Z ratios). This equivalence suggests that the Z normalization as well as Z ratio calculations adequately mirror standard mean-adjusted normalization and subsequent fold change estimates of gene expression.
|
where 226 genes were designated as up-regulated using global normalization versus 24 genes that are reported as up-regulated by the Z ratio method. Figure 3
examines this effect by illustrating the results obtained using either global normalization or Z score transformation as applied to the two separate datasets: PMA+I versus control (Figure 3 A–C)
or PMA + 28 versus control (Figure 3 D and E)
. The top panels show the original and normalized data for PMA+I. In this case, both normalization methods appear to balance the data symmetrically and, therefore, it is not surprising that the resulting gene changes are found to be highly similar, as previously mentioned (Figure 2)
. In contrast, the original and normalized data for PMA + 28 (Figure 2 D–F)
, demonstrate that Z score transformation but not global normalization is capable of correcting the data distortion (indicated by arrows) introduced by the PMA + 28 hybridization data. Although the exact cause for this data distortion is unclear (but neither is it uncommon; a false positive rate associated with low gene expression has been noted by other investigators20
), it clearly accounts for the skewed results reported for PMA+I up-regulation in Table 2
, mostly as the result of a bias in the data in the PMA + 28 direction at lower expression levels. Furthermore, it should also be clear that inclusion of the PMA + 28 data in any extended set of experimental comparisons would continue to exert a distorting effect. In general, Z normalization shows greater stability as a result of examining where each gene intensity falls in the overall distribution of values within a given array as opposed to adjusting all of the genes in an array by a single common value.
|
|
. Z scores allow clustering and other analytical methods to be used on corrected values that are directly related to original intensity values as well as on comparative ratios. The Z scores used here were generated from a wider range of experiments than from the study of T cell activation previously shown and include data from three independent human T-cell donors as well as from cultured Jurkat T cells. The unsupervised clustering algorithm (both genes and experiments were allowed to cluster independently) clearly separates not only the two distinct cell types (Jurkat and donor T cells) but also precisely sorts the primary T-cell cultures by individual donor. The high correlation of clustered experimental groups to the original experimental identities suggests that Z score transformation by itself is a reliable normalization process, providing a basic unit of comparison between a wide range of samples being analyzed (within a single microarray type).
|
. The data shown here is a sampling of the most significant genes up-regulated by treatment with either PMA+I (A) or PMA + 28 (B). The majority of all significant changes in gene expression as calculated by any of the tested methods can be seen in Figure 5
. Gene changes exceeding specified significance thresholds are shaded in gray. The Z test, Z ratio, fold change, and SAM analysis data in this example were taken from the same set of experiments using T cells harvested from a single donor. These results were then compared to a composite of replicated data from three individual donors (all donors Z test) by which the data are sorted in descending order.
|
shows that the fold changes calculated from globally normalized data appear to greatly underestimate the number of significant gene changes as predicted by all other methods. The Z test and Z ratio results appear to agree quite well with each other with some exceptions. The Z test was somewhat more conservative than the Z ratios for determining significant changes in gene expression. Both Z ratios and Z tests overlap with the significance determinations made by the SAM program. In addition, and perhaps, most importantly, the Z ratio, Z test and SAM predictions made from data from a single donor appear to reliably mirror the significance estimates made by performing the Z test on the combined experimental data derived from three individual donors. The significant overlap between three independent statistical analysis methods and the agreement between single and multiple donor data suggests that, in the aggregate, reliable changes in gene expression can be consistently detected by all three methods. |
Discussion |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
Z scores provide a relative, semiquantitative estimate of gene expression levels and, as such, form the basis of comparison of hybridization intensity data among many experiments within the same array type. Direct inspection of Z score values in visualization analyses such as hierarchical clustering is aided by the fact that Z scores are proportional to the intensity of the original hybridization signal. The value of the Z score is directly reflective of the underlying differential hybridization values (ie, higher positive Z scores represent the most highly expressed genes, lower negative Z scores represent the least expressed genes). Thus Z scores provide a useful and intuitive method for visualizing and interpreting very large amounts of data in their natural biological context. This is in contrast to normalization strategies that express hybridization intensities as ratios of one sample to another (either experimental or to a common reference sample). The values derived by ratio normalization techniques are more difficult to interpret because they are always dependent on the normalizing sample from which they were derived. Positive and negative values in these analyses simply indicate their relationship to the normalizing sample rather than reflecting actual gene expression levels. Ratio normalization thus makes it difficult to compare many different experiments directly even when using the same array type.
Z ratios provide a relative measure of significant gene expression changes in pair-wise group comparisons. In this regard, Z ratios are the conceptual equivalent to Cy3/Cy5 ratios generated using two-color fluorescent techniques. Just as Z scores are used to analyze many different experiments in terms of relative intensity measurements, Z ratios can be used to compare significant changes in gene expression across a similarly wide range of experiments. The advantages of Z ratios are that they are directly comparable among many different experiments, rapidly calculated, and show good agreement (Figure 5)
with more complex statistical analyses (eg, SAM analysis).
The application of Z test statistics to Z score microarray data was used to address additional requirements for a more rigorous statistical analysis than provided for solely by Z ratios. These improvements include, in the Z test, a SE method for balancing the effects of repeated measurement variation versus the statistical power afforded by replicate numbers. Because the Z test places a high value on low variability between experimental replicates, it tends to be more conservative than Z ratios for finding significant gene changes. Indeed, the impact on significance calculations of sample variation when using the Z test is such that it has mitigated the need, in our hands, for a priori outlier removal. The use of the Z test is facilitated directly in the Excel program, which provides one-tailed P values for the Z distribution (function = NORMSDIST).
The comparability of data are critical in the handling of the ever-increasing data streams being generated during ongoing microarray gene expression studies. Z scores will not, theoretically, be comparable outside of the array type from which they are generated since their value is specifically linked to a fixed population of genes. This limitation, however, is almost universal in the field of microarray techniques making cross-array and cross-platform comparisons difficult and inhibiting the growth of universal databases. A reliable method such as Z score transformation is, nonetheless, vital for intra-array comparisons in large studies using a stable focused array format (eg, NIA-Immunoarray).
|
Acknowledgments |
|---|
|
Footnotes |
|---|
Supported by the William Lion Penzner Foundation (MPV).
C. C. and M.P.V. contributed equally to this work.
Accepted for publication December 16, 2002.
|
References |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
This article has been cited by other articles:
 |
|
 |
|
  M. J. Mattapallil, A. Augello, C. Cheadle, D. Teichberg, K. G. Becker, C.-C. Chan, J. J. Mattapallil, G. Pennesi, and R. R. Caspi Differentially Expressed Genes in MHC-Compatible Rat Strains That Are Susceptible or Resistant to Experimental Autoimmune Uveitis Invest. Ophthalmol. Vis. Sci., May 1, 2008; 49(5): 1957 - 1970. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 J. Walston, N. Fedarko, H. Yang, S. Leng, B. Beamer, S. Espinoza, A. Lipton, H. Zheng, and K. Becker The Physical and Biological Characterization of a Frail Mouse Model J. Gerontol. A Biol. Sci. Med. Sci., April 1, 2008; 63(4): 391 - 398. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 M. Eszlinger, K. Krohn, A. Kukulska, B. Jarzab, and R. Paschke Perspectives and Limitations of Microarray-Based Gene Expression Profiling of Thyroid Tumors Endocr. Rev., May 1, 2007; 28(3): 322 - 338. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 G. P. Sawiris, K. G. Becker, E. J. Elliott, R. Moulden, and R. G. Rohwer Molecular analysis of bovine spongiform encephalopathy infection by cDNA arrays J. Gen. Virol., April 1, 2007; 88(4): 1356 - 1362. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 B J M Braakhuis, A Senft, R de Bree, J de Vries, B Ylstra, J Cloos, D J Kuik, C R Leemans, and R H Brakenhoff Expression profiling and prediction of distant metastases in head and neck squamous cell carcinoma J. Clin. Pathol., December 1, 2006; 59(12): 1254 - 1260. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 M. Dolled-Filhart, L. Ryden, M. Cregger, K. Jirstrom, M. Harigopal, R. L. Camp, and D. L. Rimm Classification of breast cancer using genetic algorithms and tissue microarrays. Clin. Cancer Res., November 1, 2006; 12(21): 6459 - 6468. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 A. E. Teschendorff, A. Naderi, N. L. Barbosa-Morais, and C. Caldas PACK: Profile Analysis using Clustering and Kurtosis to find molecular classifiers in cancer Bioinformatics, September 15, 2006; 22(18): 2269 - 2275. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 S. Penfield, Y. Li, A. D. Gilday, S. Graham, and I. A. Graham Arabidopsis ABA INSENSITIVE4 Regulates Lipid Mobilization in the Embryo and Reveals Repression of Seed Germination by the Endosperm PLANT CELL, August 1, 2006; 18(8): 1887 - 1899. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 N. Harish, R. Gupta, P. Agarwal, V. Scaria, and B. Pillai DyNAVacS: an integrative tool for optimized DNA vaccine design. Nucleic Acids Res., July 1, 2006; 34(Web Server issue): W264 - W266. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
J.-i. Kawada, H. Kimura, Y. Kamachi, K. Nishikawa, M. Taniguchi, K. Nagaoka, H. Kurahashi, S. Kojima, and T. Morishima Analysis of gene-expression profiles by oligonucleotide microarray in children with influenza J. Gen. Virol., June 1, 2006; 87(6): 1677 - 1683. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 R. Stepanyan, K. Day, J. Urban, D. L. Hardin, R. S. Shetty, C. D. Derby, B. W. Ache, and T. S. McClintock Gene expression and specificity in the mature zone of the lobster olfactory organ. Physiol Genomics, April 13, 2006; 25(2): 224 - 233. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
J. Fan, M. Zhan, J. Shen, J. L. Martindale, X. Yang, T. Kawai, and M. Gorospe En masse nascent transcription analysis to elucidate regulatory transcription factors Nucleic Acids Res., March 15, 2006; 34(5): 1492 - 1500. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 T. W. Plaia, R. Josephson, Y. Liu, X. Zeng, C. Ording, A. Toumadje, S. N. Brimble, E. S. Sherrer, E. W. Uhl, W. J. Freed, et al. Characterization of a New NIH-Registered Variant Human Embryonic Stem Cell Line, BG01V: A Tool for Human Embryonic Stem Cell Research Stem Cells, March 1, 2006; 24(3): 531 - 546. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 J. Cloos, W. P.H. de Boer, M. H.J. Snel, P. van den IJssel, B. Ylstra, C. R. Leemans, R. H. Brakenhoff, and B. J.M. Braakhuis Microarray Analysis of Bleomycin-Exposed Lymphoblastoid Cells for Identifying Cancer Susceptibility Genes Mol. Cancer Res., February 1, 2006; 4(2): 71 - 77. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 I. Lopez de Silanes, S. Galban, J. L. Martindale, X. Yang, K. Mazan-Mamczarz, F. E. Indig, G. Falco, M. Zhan, and M. Gorospe Identification and Functional Outcome of mRNAs Associated with RNA-Binding Protein TIA-1 Mol. Cell. Biol., November 1, 2005; 25(21): 9520 - 9531. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 S. F. Siddiqui, J. Pawelek, T. Handerson, C.-Y. Lin, R. B. Dickson, D. L. Rimm, and R. L. Camp Coexpression of {beta}1,6-N-Acetylglucosaminyltransferase V Glycoprotein Substrates Defines Aggressive Breast Cancers with Poor Outcome Cancer Epidemiol. Biomarkers Prev., November 1, 2005; 14(11): 2517 - 2523. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 J. J. Hoefnagel, R. Dijkman, K. Basso, P. M. Jansen, C. Hallermann, R. Willemze, C. P. Tensen, and M. H. Vermeer Distinct types of primary cutaneous large B-cell lymphoma identified by gene expression profiling Blood, May 1, 2005; 105(9): 3671 - 3678. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
N. Strunnikova, C. Zhang, D. Teichberg, S. W. Cousins, J. Baffi, K. G. Becker, and K. G. Csaky Survival of Retinal Pigment Epithelium after Exposure to Prolonged Oxidative Injury: A Detailed Gene Expression and Cellular Analysis Invest. Ophthalmol. Vis. Sci., October 1, 2004; 45(10): 3767 - 3777. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
S. Zareparsi, A. Hero, D. J. Zack, R. W. Williams, and A. Swaroop Seeing the Unseen: Microarray-Based Gene Expression Profiling in Vision Invest. Ophthalmol. Vis. Sci., August 1, 2004; 45(8): 2457 - 2462. [Full Text] [PDF]
|
|
|
|
|
|
T. Kawai, J. Fan, K. Mazan-Mamczarz, and M. Gorospe Global mRNA Stabilization Preferentially Linked to Translational Repression during the Endoplasmic Reticulum Stress Response Mol. Cell. Biol., August 1, 2004; 24(15): 6773 - 6787. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 I. L. de Silanes, M. Zhan, A. Lal, X. Yang, and M. Gorospe Identification of a target RNA motif for RNA-binding protein HuR PNAS, March 2, 2004; 101(9): 2987 - 2992. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 M. A. Rubin, M. P. Zerkowski, R. L. Camp, R. Kuefer, M. D. Hofer, A. M. Chinnaiyan, and D. L. Rimm Quantitative Determination of Expression of the Prostate Cancer Protein {alpha}-Methylacyl-CoA Racemase Using Automated Quantitative Analysis (AQUA): A Novel Paradigm for Automated and Continuous Biomarker Measurements Am. J. Pathol., March 1, 2004; 164(3): 831 - 840. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
S. J. Hong, H. Li, K. G. Becker, V. L. Dawson, and T. M. Dawson Identification and analysis of plasticity-induced late-response genes PNAS, February 17, 2004; 101(7): 2145 - 2150. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 G. ERMAK, C. CHEADLE, K. G. BECKER, C. D. HARRIS, and K. J. A. DAVIES DSCR1(Adapt78) modulates expression of SOD1 FASEB J, January 1, 2004; 18(1): 62 - 69. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||
2, for Z ratios at 
0.01, for SAM at d > 1.45. SAM, fold changes, Z ratios, and Z test data are from a single donor. Labeling replicates (3 for the control sample RNA, 2 each for PMA+I and PMA + 28 sample RNAs) were first averaged for Z ratio and fold change estimates. These results are compared to a composite of replicated data (using averages of multiple individual donor RNA labelings) from three individuals (all donors Z test). The data are sorted in order of decreasing Z test value for the all donors column.