Accuracy of cDNA microarray methods to detect small gene expression changes induced by neuregulin on breast epithelial cells
© Yao et al; licensee BioMed Central Ltd. 2004
Received: 19 May 2004
Accepted: 23 July 2004
Published: 23 July 2004
cDNA microarrays are a powerful means to screen for biologically relevant gene expression changes, but are often limited by their ability to detect small changes accurately due to "noise" from random and systematic errors. While experimental designs and statistical analysis methods have been proposed to reduce these errors, few studies have tested their accuracy and ability to identify small, but biologically important, changes. Here, we have compared two cDNA microarray experimental design methods with northern blot confirmation to reveal changes in gene expression that could contribute to the early antiproliferative effects of neuregulin on MCF10AT human breast epithelial cells.
We performed parallel experiments on identical samples using a dye-swap design with ANOVA and an experimental design that excludes systematic biases by "correcting" experimental/control hybridization ratios with control/control hybridizations on a spot-by-spot basis. We refer to this approach as the "control correction method" (CCM). Using replicate arrays, we identified a decrease in proliferation genes and an increase in differentiation genes. Using an arbitrary cut-off of 1.7-fold and p values <0.05, we identified a total of 32 differentially expressed genes, 9 with the dye-swap method, 18 with the CCM, and 5 genes with both methods. 23 of these 32 genes were subsequently verified by northern blotting. Most of these were <2-fold changes. While the dye-swap method (using either ANOVA or Bayesian analysis) detected a smaller number of genes (14–16) compared to the CCM (46), it was more accurate (89–92% vs. 75%). Compared to the northern blot results, for most genes, the microarray results underestimated the fold change, implicating the importance of detecting these small changes.
We validated two experimental design paradigms for cDNA microarray experiments capable of detecting small (<2-fold) changes in gene expression with excellent fidelity that revealed potentially important genes associated with the anti-proliferative effects of neuregulin on MCF10AT breast epithelial cells.
Spotted cDNA microarrays are used in high-throughput experiments that interrogate the relative expression of thousands of genes simultaneously for many biological processes with wide applications in biological and medical research. Typically in a two-dye spotted cDNA microarray experiment, two mRNA samples are transcribed into cDNAs, labeled with two different fluorescent dyes, commonly Cy3 and Cy5, and hybridized on the same slide. The relative gene expression level is then measured as a ratio of the intensities of the fluorescent dyes. However, the signal intensity of the dye, which indirectly represents the gene expression level, can be affected by many other sources of error such as dye efficiency, sample preparation, and the variability of the biological samples [1, 2].
An important question is how to identify differentially expressed genes, some of which change only minimally (<2-fold), given many known and potentially unknown sources of variance in the microarray experiment. In order to reduce false positive rates, many published experiments use a cut-off of 2- to 3-fold [3–5]. This limits the ability of the microarray experiment to detect small, but biologically important changes. In fact, recent reports have shown that microarrays can significantly underestimate gene expression changes and therefore a high cut-off will miss important changes . Although more sophisticated statistical methods have been proposed for single slide analysis [7–13], it is becoming clear that in order to reduce random variance, replication becomes more and more important in microarray experimental design by greatly increasing the power of the experiment to measure small gene expression changes [2, 13–17]. As a relatively new technique, many new theories have been developed for data analysis and experimental design, but few of these theories have been rigorously tested against a well-established standard method such as the Northern blot.
In this paper we compared two experimental design and analysis methods performed on quadruplicate arrays that include a dye-swap design [18, 19] and a modified reference design method that uses a control-control hybridization to correct for systematic experimental errors, that we refer to as the "control correction method" (CCM). We demonstrate that both experimental designs accurately identified small (<2-fold) gene expression changes after a 24-hour treatment of MCF10AT breast epithelial cells with the growth and differentiation factor neuregulin. These changes correlate well with the anti-proliferative effects of neuregulin resulting in a relative decrease in proliferative genes and increase in anti-proliferative genes that will be important for future investigations.
Experimental designs to address systematic errors
Given the unavoidable presence of these systematic errors, methods to correct these errors are needed. One way to correct for systematic errors in microarray experiments is to take advantage of C/C hybridizations to correct the T/C hybridizations. This requires a modified reference design, which we refer to as a "control correction" design. This is different from a common reference design used previously [19, 20]. Here, each spot of the T/C hybridization is "corrected" by the same spot from the C/C hybridization for systematic errors. A second method that will also correct for systematic errors is a "dye-swap" design [16, 17, 19]. The dye-swap design uses an ANOVA to calculate gene expression changes from replicate cDNA microarrays probed with T/C hybridizations performed where the dye color is swapped. Included in the ANOVA are factors to correct for systematic errors such as dye and dye-gene interactions. The "control correction" and the dye-swap designs are compared in Fig. 2B. Each of these experimental designs was performed on quadruplicate arrays. Each of these two designs required its own analysis method. While we used an analysis method that utilizes individual t-tests for each spot for the CCM, we compared both ANOVA and Bayesian analysis methods for the dye-swap design.
Control correction method experimental design and results
Comparison of the control correction to the dye-swap design
Many have proposed that a dye-swap experimental design combined with an ANOVA will correct for systematic errors [17–19]. To verify this and compare the dye-swap design to the control correction design, a dye-swap experiment was performed on quadruplicate arrays using the same RNA samples and the two interconnect ANOVA model of Wolfinger et al [22, 23]. Using this experimental design with the same cut-off values, 14 differentially expressed genes were identified and are presented as a volcano plot alongside that of the CCM (Fig. 6B).
List of identified genes. Gene accession numbers, gene descriptions, fold-changes, and p-values for genes identified by the dye-swap method with ANOVA and regularized t-test analysis and the control correction method (CCM). Genes are broadly classified into three groups: proliferation-related, differentiation-related and unclassified.
Heat shock proteins
heat shock 70 kDa protein 8
heat shock 60 kDa protein 1 (chaperonin)
heat shock 70 kDa protein 9B (mortalin-2)
Human mitochondrial matrix protein P1
Human calnexin mRNA, complete cds
Human chaperonin protein (Tcp20) gene complete cds
Transcription and translation
helicase with zinc finger domain
eukaryotic translation initiation factor 4A, isoform 2
Human translation repressor NAT1 mRNA, complete cds
transcription factor 4
Human mRNA for B-myb gene
Human SREBP-1 mRNA
eukaryotic translation elongation factor 1 alpha 1-like 14
Fatty acid and sugar metabolism
Homo sapiens mRNA for mitochondrial 3-ketoacyl-CoA thiolase beta-subunit of trifunctional protein, complete cds
Human mRNA for lactate dehydrogenase B (LDH-B)
Homo sapiens mRNA for squalene epoxidase, complete cds
Human glutamate dehydrogenase (GDH) mRNA, complete cds
3-oxoacid CoA transferase
Homo sapiens mRNA for stearoyl-CoA desaturase
Homo sapiens mRNA for GDP dissociation inhibitor beta
Human mRNA for n-chimaerin
Human cyclin G1 mRNA, complete cds
Homo sapiens BASS1 (BASS1) mRNA, partial cds
Homo sapiens mRNA for cytochrome c, partial cds
H. sapiens mRNA for S-adenosylmethionine synthetase
Homo sapiens neuronal apoptosis inhibitory protein mRNA, complete cds
alpha B-crystallin=Rosenthal fiber component [human, glioma cell line, mRNA, 691 nt]
H. sapiens mRNA for FAST kinase
Human mRNA of trk oncogene
Human fgr proto-oncogene encoded p55-c-fgr protein, complete cds
H. sapiens cDNA for RFG
platelet-activating factor acetylhydrolase, isoform Ib, beta subunit 30 kDa
Human branched-chain amino acid aminotransferase (ECA40) mRNA, complete cds
Human thrombospondin 2 (THBS2) mRNA, complete cds
Human mRNA for receptor of retinoic acid
Human mRNA for serine protease
Homo sapiens lung cancer antigen NY-LU-12 variant A mRNA, complete cds
Human small proline rich protein (sprI) mRNA, clone 15B
ECM and vesicle trafficking
Human leukocyte adhesion protein (LFA-1/Mac-1/p150,95 family) beta subunit mRNA
H4(D10S170) = putative cytoskeletal protein [human, thyroid, mRNA, 3011 nt]
SNAP25 synaptosomal-associated protein, 25 kDa
Human thrombospondin 2 (THBS2) mRNA, complete cds
H. sapiens h-Sp1 mRNA
collagen, type I, alpha 1
chemokine (C-C motif) receptor 6
Human mRNA for complement component C1r
ring finger protein 144
Homo sapiens mRNA for DNA polymerase beta, complete cds
myosin, light polypeptide 9, regulatory
Human mRNA for erythrocyte adducin alpha subunit
Homo sapiens mRNA for putative methyltransferase
Human mRNA for 27-kDa calbindin
aldehyde dehydrogenase 4 family, member A1
Human nicotinic acetylcholine receptor alpha3 subunit precursor, mRNA, complete cds
Human clone 23945 mRNA, complete cds
Homo sapiens dystrophin (DMD) mRNA, complete cds
Human sarcomeric mitochondrial creatine kinase(MtCK) gene, complete cds
Human mRNA for lactoyl glutathione lyase
Homo sapiens mRNA for p27
Human ribosomal protein mRNA
Verification of microarray accuracy by northern blot analysis
Since the ANOVA method we used can sometimes underestimate the variance, we re-analyzed our dye-swap data with a Bayesian method using a regularized t-test as implemented in Cyber-T . This analysis revealed 16 differentially expressed genes using the same cut-offs, 10 of which were in common with the ANOVA method (Table 1). A greater number of genes were identified using the regularized t-test, and the corresponding p values for these genes were lower. Based on the previous northern blot data, 8/9 (89%) of these were confirmed.
Gene expression changes in MCF10AT cells suggest a rapid anti-proliferative effect of neuregulin
MCF10AT cells are a human breast epithelial cell line stably transfected with a mutant ras oncogene. These cells are pre-malignant, but can progress to invasive carcinoma [25, 26]. Given that neuregulin can differentially affect the growth properties of different cell lines, we used the MCF10AT cell line as model system to identify genes that may be down-stream from neuregulin activation and could thus be studied further for their roles in breast cancer cells that respond differentially to neuregulin. Combining two cDNA microarray experimental design methods, we have identified genes differentially expressed by neuregulin treatment that correlated with a significant decrease in their growth rate. The pattern of expression clearly shows an anti-proliferative effect of neuregulin on the MCF10AT cells with a reduction in genes associated with proliferation such as heat shock proteins, oncogenes, cell cycle control genes, genes involved in fatty acid and sugar synthesis, transcription and translation together with an increase in differentiation genes including tumor suppressor genes, DNA damage repair genes, growth inhibition genes and differentiation genes. We further showed that these effects are biologically consistent with the rapid, anti-proliferative effects of neuregulin on cell number. Additional experiments have shown that these genes are important biological markers for the degree of malignancy in other breast epithelial cell lines that have differential proliferation responses to neuregulin (Li Q, Ahmed S, and Loeb JA, unpublished results).
Both experimental designs demonstrate a high confirmation rate for small changes in gene expression
One of the important tasks in microarray technology is to design experiments and develop statistical tools to obtain data efficiently and accurately to answer fundamental questions in biology. In many experiments, this requires the ability to detect small changes in gene expression with high fidelity. In this study we compared two common experimental design paradigms for cDNA microarrays and determined their accuracy by northern blot. Both methods identified small expression changes with considerable accuracy. In the control correction design, we used control hybridizations to correct for systematic errors on a spot-by-spot basis. The method is based on an assumption that systematic errors from slides made from the same lot and processed identically do not vary significantly. To minimize the possible variance of systematic errors in T/C slides and C/C slides we maintained strict experimental conditions, such as same-day sample preparation and same-day hybridization. We also used the same control samples for both the T/C and C/C hybridizations instead of using an arbitrary control sample that might be quite different in mRNA composition . This results in similar spot intensities for each gene both in the treatment and the control and will minimize any differences that could be caused by the different mRNA compositions from different samples. This spot-by-spot control correction can eliminate systematic errors that cannot be corrected with slide-wise normalization. Similarly, in the dye-swap design, two different dyes are used to label the same sample, which enables the correction of dye-gene interactions in the ANOVA model.
The t-test used for the CCM and ANOVA for the dye-swap method depend on assumptions of Gaussian distributions that may or may not be present in a microarray experiment with a small number of replicates. Some efforts have been made to develop Bayesian frameworks that incorporate prior distributions in order to estimate the noise [24, 29, 30]. We therefore re-analyzed our dye-swap data using a "regularized" t-test . Using this, we identified 16 genes that met our cut-off criteria, 10 of which were in common with the ANOVA analysis. Of those genes that we measured by northern blot analysis, 8/9 or 89% were verified. In summary, the regularized t-test revealed more genes than the ANOVA method with generally lower p values. If we eliminate the 1.7-fold cut-off, but maintain the p value <0.05, the CCM identified 493 genes, the ANOVA identified 499 genes, and the regularized t-test identified 729 differentially expressed genes (Fig. 8B). Among these, 399 were in common between the regularized t-test and ANOVA, 248 in common between the CCM and the regularized t-test, and 188 in common between the CCM and the ANOVA. These results demonstrate that if the false-positive rate remains the same, the regularized t-test is more sensitive than the traditional ANOVA and has extensive overlap, while the CCM has the least overlap between the other methods, but identifies different genes with slightly less specificity.
In our analysis, we selected genes based on their p values obtained from replicates of individual spots and did not adjust these p-values for multiple comparisons. This may be a major cause for the higher false positive rates for both of our experimental designs. For the CCM, if we apply Bonferroni correction, while we can eliminate all false positives, we would also miss a majority of the differentially expressed genes verified by Northern blotting. Therefore, if accuracy is the main purpose of a study, multiple comparison corrections should be used, while if sensitivity is the main purpose, then it should not be used with the understanding that the accuracy will be lower.
Comparison of a dye-swap versus a control correction method experimental design
For our experimental design, the dye-swap method had a higher confirmation rate than the control correction method. This is, in part, due to the smaller variance that results from an effective doubling of the number of treated samples in the dye-swap method compared to the control correction method. Despite the higher degree of accuracy, the dye-swap design identified fewer genes and only detected down-regulated genes, whereas the control correction identified 3-times the number of genes that were both up- and down-regulated. However the control correction method was less specific for up-regulated genes. These differences may not solely reflect methodological differences, but likely result from experimental variability produced by performing the experiments independently on different days. Nonetheless, the results presented here suggest that both methods have clear merit in their abilities to show true gene expression changes, particularly for expression changes of 2-fold or less, and for genes with low signal intensities and/or low abundance.
The final decision as to which method is preferred depends on the experimental design. For example, the amount of sample and number of replicates required are important considerations both in terms of how difficult the RNA is to obtain and the number of samples that need to be compared. This also translates into the cost to perform the experiment. For instance, the dye-swap method generates a larger sample size for the same number of slides, thus producing greater significance when comparing gene expression between two samples. However this method requires a minimum of two slides and two different labeling reactions per sample. If the amount of sample is limited or population level replication is more desirable than individual sample replication, the control correction is more efficient since individual replicates for reverse dye labeling are not required and each sample can be run with only one slide. For example, to compare 6 treatment samples with a single control sample would require a minimum of 12 microarrays using the dye-swap method, whereas the minimum number of 8 arrays is possible using the control correction method; 6 for treatment samples and 2 for controls.
Another common experimental design used for time course or dose response studies is the reference design. In fact, the control correction method described here is essentially a modified reference design method where the zero time or dose point is the control-control comparison. As discussed above, using a very similar control sample to correct the series will give less false positives and negatives and a more accurate absolute value of the observed change than a dissimilar, pooled reference sample.
Under-estimation of fold changes by cDNA microarrrays
Although our cDNA microarray results were accurate, the measured changes generally underestimated the actual changes measured by northern blots. Yuen et al.  similarly found that both oligonucleotide arrays (GeneChips by Affymetrix) and cDNA arrays underestimate fold changes compared to quantitative RT-PCR. The cause for this underestimation is not clear, however, it may be due to the limited dynamic range of dye signal or non-specific binding of the dye. Nonetheless, the limitations in accuracy and fold change estimation are far outweighed by the ability of microarrays to identify biologically important gene expression changes.
This study demonstrated that dye-swap and control correction experimental design paradigms for cDNA microarray experiments are capable of detecting small, biologically important changes in gene expression with excellent fidelity while revealing important down-stream anti-proliferative effects of neuregulin on breast epithelial cells for future studies.
Human cDNA glass microarrays, called the Alphamax Genechip, were obtained from Alphagene Inc. (Woburn, MA) containing 3333 cDNAs spotted in duplicate. The cDNAs used are identical to commercially available Micromax 2400 slides from Perkin Elmer Life Sciences (Boston, MA), most of which were derived from a human fetal brain cDNA library, with an additional 933 genes (gene list available upon request).
MCF10AT cell culture – MCF10AT cells were from Dr. Robert Pauley at the Karmanos Cancer Institute (Detroit, MI). The cells were cultured in DMEM/F12 media (Invitrogen) supplemented with 5% horse serum (Invitrogen), 10 mM HEPES buffer (Invitrogen), 10 μg/ml insulin (Sigma), 20 ng/ml EGF (Upstate Biotechnology), 100 ng/ml cholera enterotoxin (CalBiochem) and 0.5 μg/ml hydrocortisone (Sigma) at 37°C in 5% CO2 incubator.
Neuregulin treatment and RNA extraction – A recombinant human NRG β1 polypeptide (amino acids 14–246) was generously provided by AMGEN (Thousand Oaks, CA). After 3 days of culture, MCF10AT cells were treated with human recombinant neuregulin β1 form for 24 hours. MCF10AT cells grown under similar conditions without neuregulin treatment were used as a control. The cells were then harvested and total RNA was extracted using Ultraspec (Biotecx laboratories). The total RNA was cleaned up by Rneasy kit (Qiagen) and quantified using a fluorescent dye binding assay, Ribogreen (Molecular Probes). RNA purity was assessed by agarose gel electrophoresis. Proliferation assays were performed by counting quadruplicate cultures plated at 5000 cells/well using a hemocytometer.
Microarray hybridization – cDNA microarrays were used in a 2-step hybridization protocol that was optimized for the Genisphere dendrimer labeling method. Total RNA was reverse transcribed into cDNA containing a unique 5' primer tag, using the Genisphere 3DNA expression array detection kit. In brief, for each reaction, 3 μg of total RNA was reverse transcribed using 0.2 μM oligo-dT-Genisphere capture primer, 0.5 mM dNTP, 200 U Superscript II (Invitrogen) in 1X first strand Superscript II buffer at 42°C for 2 h. The RNA from the DNA/RNA hybrids was denatured with 0.5 M NaOH / 50 mM EDTA at 65°C for 10 min. The reaction was neutralized using 1 M Tris-HCl ph 7.5. The contents of the tube containing the NRG-treated and control cDNA were then mixed together and 3 μl linear acrylamide (Ambion) and 250 μl of 3 M ammonium acetate were added to them. cDNA was precipitated by adding 100% Ethanol and incubating at -20°C for 30 min. The cDNA was collected in a pellet by centrifugation at 13000 rpm for 15 min in a microcentrifuge. and resuspended in Alternate (formamide-containing) Hybridization buffer (Genisphere) at 65°C for 10 min and modified LNA blocker (Genisphere) with denatured Cot 1 DNA.
The entire mixture was added to the pre-hybridized array (Alphamax) for hybridization at 55°C for at 36 hr. A clear increase in signal was obtained with a 36 h hybridization compared to 16 h. After hybridization, the arrays were washed with 2X SSC and 0.2% SDS at 60°C for 15 min, followed by a wash with 2X SSC and another with 0.2X SSC at room temperature. For fluorescence detection, a second hybridization with the dendrimer was optimal. 2.5 μl each of the Cy3 and Cy5 dendrimer in Hybridization Buffer (Vial 6, Genisphere kit) were mixed with denatured Cot1 DNA and differential expander and the mixture was added to the pre-hybridized slides for hybridization at 60°C for 2 hrs. The slides were washed again as described above.
Microarray data analysis method
Analysis of CCM experiment
Arrays were scanned with a GenePix 4000 A scanner (Axon Instruments, Inc., Union City, CA). Images were quantified using ImaGene Software (Biodiscovery, Inc. Marina del Rey, CA) that uses a local background subtracted from the signal. Signals not consistently detectable (background corrected signal lower than 2 times of background standard deviation) were eliminated.
We fitted loess curve to the log transformed data using the "loess" function in SAS software (SAS Institute Inc., NC) for intensity dependent normalization followed by a t-test to compare T/C with C/C ratio, gene by gene. The t-test was performed on the normalized log ratio with Welch correction for unequal variance. The control corrected fold change was calculated as:
log (fold) = log(T/C)-log(C/C)
Analysis of dye-swap experiment
For the dye-swap method we performed the same background correction and data filtering for absent genes and log transformations. We then used a two interconnect ANOVA model [22, 23] and Mixed Model Analysis of Microarray Data (MANMADA) http://statgen.ncsu.edu/ggibson/Manual.htm to identify differentially expressed genes. First we use a normalization model for log-transformed intensity measurements:
y ij = μ + A i + D j + AD ij + ε ij
Where μ is the sample mean, A i is the effect of i th array, D j is the effect of dye cy3 or cy5, AD ij is array dye interaction and ε ij is random error. The residue from normalization model is then used in following gene model to find treatment effects on each gene:
r ijkg = A ig + D jg + T kg
Where r ijkg is the residual of each gene from the normalization model, T kg is the treatment effect (control or treated), and A ig and D jg are the array and dye effects, respectively. The expression change for each gene is thus:
log (fold) = Ttreated-Tcontrol
5 μg total RNA isolated from MCF10AT cells was run on a 1.3% Agarose/2.2M Formaldehyde gel as described previously . Probes were prepared by PCR from the same clones used to spot the slides provided by Alphagene Inc except for AJ224442, X86779 and U62739, where clones BC011696, BI754516 and BG763631, with of over 99% identity, were used as substitutes. Probes were generated by random priming using PrimiT II kit (Stratagene) radiolabeled probes. The auto-radiographs within the linear range of the film were scanned with a flatbed scanner with transparency adapter and quantified using MetaMorph (Universal Imaging) analysis software as described previously . For time course measurements, the amount of signal normalized for loading with either 18S RNA or GAPDH were plotted together after first setting 100% to the intensity of the control measurement at 48 hours and setting the lowest intensity value to 0%.
This work was supported by the Ralph C. Wilson, Sr. and Ralph C. Wilson, Jr. Medical Research Foundation (JAL), NINDS (NIH) R01 NS45207 (JAL), and the American Cancer Society 85-003-14 (JAL). SNR was supported by a pre-doctoral fellowship from the Epilepsy Foundation of America. We thank Robert Getts from Genisphere for helpful discussions on optimizing the dendrimer labeling system and Thomas Beaumont for helpful comments on the manuscript.
- Nadon R, Shoemaker J: Statistical issues with microarrays: processing and analysis. Trends Genet 2002, 18(5):265–271. 10.1016/S0168-9525(02)02665-3View ArticlePubMedGoogle Scholar
- Draghici S, Kuklin A, Hoff B, Shams S: Experimental design, analysis of variance and slide quality assessment in gene expression arrays. Curr Opin Drug Discov Devel 2001, 4(3):332–337.PubMedGoogle Scholar
- Schena M, Shalon D, Heller R, Chai A, Brown PO, Davis RW: Parallel human genome analysis: microarray-based expression monitoring of 1000 genes. Proc Natl Acad Sci U S A 1996, 93(20):10614–10619. 10.1073/pnas.93.20.10614PubMed CentralView ArticlePubMedGoogle Scholar
- Brudler R, Hitomi K, Daiyasu H, Toh H, Kucho K, Ishiura M, Kanehisa M, Roberts VA, Todo T, Tainer JA, Getzoff ED: Identification of a new cryptochrome class. Structure, function, and evolution. Mol Cell 2003, 11(1):59–67. 10.1016/S1097-2765(03)00008-XView ArticlePubMedGoogle Scholar
- DeRisi JL, Iyer VR, Brown PO: Exploring the metabolic and genetic control of gene expression on a genomic scale. Science 1997, 278(5338):680–686. 10.1126/science.278.5338.680View ArticlePubMedGoogle Scholar
- Yuen T, Wurmbach E, Pfeffer RL, Ebersole BJ, Sealfon SC: Accuracy and calibration of commercial oligonucleotide and custom cDNA microarrays. Nucleic Acids Res 2002, 30(10):e48. 10.1093/nar/30.10.e48PubMed CentralView ArticlePubMedGoogle Scholar
- Newton MA, Kendziorski CM, Richmond CS, Blattner FR, Tsui KW: On differential variability of expression ratios: improving statistical inference about gene expression changes from microarray data. J Comput Biol 2001, 8(1):37–52. 10.1089/106652701300099074View ArticlePubMedGoogle Scholar
- Chen Y, Kamat V, Dougherty ER, Bittner ML, Meltzer PS, Trent JM: Ratio statistics of gene expression levels and applications to microarray data analysis. Bioinformatics 2002, 18(9):1207–1215. 10.1093/bioinformatics/18.9.1207View ArticlePubMedGoogle Scholar
- Efron B, Tibshirani R: Empirical bayes methods and false discovery rates for microarrays. Genet Epidemiol 2002, 23(1):70–86. 10.1002/gepi.1124View ArticlePubMedGoogle Scholar
- Hughes TR, Marton MJ, Jones AR, Roberts CJ, Stoughton R, Armour CD, Bennett HA, Coffey E, Dai H, He YD, Kidd MJ, King AM, Meyer MR, Slade D, Lum PY, Stepaniants SB, Shoemaker DD, Gachotte D, Chakraburtty K, Simon J, Bard M, Friend SH: Functional discovery via a compendium of expression profiles. Cell 2000, 102(1):109–126. 10.1016/S0092-8674(00)00015-5View ArticlePubMedGoogle Scholar
- Sapir M, Churchill G: Estimating the posterior probability of differential gene expression from microarray data. 2000.Google Scholar
- Draghici S: Statistical intelligence: effective analysis of high-density microarray data. Drug Discov Today 2002, 7(11):S55–63. 10.1016/S1359-6446(02)02292-4View ArticlePubMedGoogle Scholar
- Draghici S: Data Analysis Tools for DNA Microarrays. Chapman and Hall/CRC Press 2003.View ArticleGoogle Scholar
- Townsend JP: Resolution of large and small differences in gene expression using models for the Bayesian analysis of gene expression levels and spotted DNA microarrays. BMC Bioinformatics 2004, 5(1):54. 10.1186/1471-2105-5-54PubMed CentralView ArticlePubMedGoogle Scholar
- Lee ML, Kuo FC, Whitmore GA, Sklar J: Importance of replication in microarray gene expression studies: statistical methods and evidence from repetitive cDNA hybridizations. Proc Natl Acad Sci U S A 2000, 97(18):9834–9839. 10.1073/pnas.97.18.9834PubMed CentralView ArticlePubMedGoogle Scholar
- Yang YH, Speed T: Design issues for cDNA microarray experiments. Nat Rev Genet 2002, 3(8):579–588.PubMedGoogle Scholar
- Churchill GA: Fundamentals of experimental design for cDNA microarrays. Nat Genet 2002, 32(Suppl):490–495. 10.1038/ng1031View ArticlePubMedGoogle Scholar
- Kerr MK, Churchill GA: Statistical design and the analysis of gene expression microarray data. Genet Res 2001, 77(2):123–128. 10.1017/S0016672301005055PubMedGoogle Scholar
- Kerr MK, Martin M, Churchill GA: Analysis of variance for gene expression microarray data. J Comput Biol 2000, 7(6):819–837. 10.1089/10665270050514954View ArticlePubMedGoogle Scholar
- Dudoit S, Yang Y, Callow M, Speed T: Statistical methods for identifying differentially expressed genes in replicated cDNA microarray experiments. 2000.Google Scholar
- Yang YH, Dudoit S, Luu P, Lin DM, Peng V, Ngai J, Speed TP: Normalization for cDNA microarray data: a robust composite method addressing single and multiple slide systematic variation. Nucleic Acids Res 2002, 30(4):e15. 10.1093/nar/30.4.e15PubMed CentralView ArticlePubMedGoogle Scholar
- Jin W, Riley RM, Wolfinger RD, White KP, Passador-Gurgel G, Gibson G: The contributions of sex, genotype and age to transcriptional variance in Drosophila melanogaster. Nat Genet 2001, 29(4):389–395. 10.1038/ng766View ArticlePubMedGoogle Scholar
- Wolfinger RD, Gibson G, Wolfinger ED, Bennett L, Hamadeh H, Bushel P, Afshari C, Paules RS: Assessing gene significance from cDNA microarray expression data via mixed models. J Comput Biol 2001, 8(6):625–637. 10.1089/106652701753307520View ArticlePubMedGoogle Scholar
- Baldi P, Long AD: A Bayesian framework for the analysis of microarray expression data: regularized t-test and statistical inferences of gene changes. Bioinformatics 2001, 17(6):509–519. 10.1093/bioinformatics/17.6.509View ArticlePubMedGoogle Scholar
- Shekhar PV, Chen ML, Werdell J, Heppner GH, Miller FR, Christman JK: Transcriptional activation of functional endogenous estrogen receptor gene expression in MCF10AT cells: a model for early breast cancer. Int J Oncol 1998, 13(5):907–915.PubMedGoogle Scholar
- Strickland LB, Dawson PJ, Santner SJ, Miller FR: Progression of premalignant MCF10AT generates heterogeneous malignant variants with characteristic histologic types and immunohistochemical markers. Breast Cancer Res Treat 2000, 64(3):235–240. 10.1023/A:1026562720218View ArticlePubMedGoogle Scholar
- Mutch DM, Berger A, Mansourian R, Rytz A, Roberts MA: The limit fold change model: A practical approach for selecting differentially expressed genes from microarray data. BMC Bioinformatics 2002, 3(1):17. 10.1186/1471-2105-3-17PubMed CentralView ArticlePubMedGoogle Scholar
- Tusher VG, Tibshirani R, Chu G: Significance analysis of microarrays applied to the ionizing radiation response. Proc Natl Acad Sci U S A 2001, 98(9):5116–5121. 10.1073/pnas.091062498PubMed CentralView ArticlePubMedGoogle Scholar
- Townsend JP, Hartl DL: Bayesian analysis of gene expression levels: statistical quantification of relative mRNA level across multiple strains or treatments. Genome Biol 2002, 3(12):RESEARCH0071. 10.1186/gb-2002-3-12-research0071PubMed CentralView ArticlePubMedGoogle Scholar
- Tseng GC, Oh MK, Rohlin L, Liao JC, Wong WH: Issues in cDNA microarray analysis: quality filtering, channel normalization, models of variations and assessment of gene effects. Nucleic Acids Res 2001, 29(12):2549–2557. 10.1093/nar/29.12.2549PubMed CentralView ArticlePubMedGoogle Scholar
- Loeb JA, Fischbach GD: Neurotrophic factors increase neuregulin expression in embryonic ventral spinal cord neurons. J Neurosci 1997, 17(4):1416–1424.PubMedGoogle Scholar
- Li Q, Loeb JA: Neuregulin-heparan-sulfate proteoglycan interactions produce sustained erbB receptor activation required for the induction of acetylcholine receptors in muscle. J Biol Chem 2001, 276(41):38068–38075.PubMedGoogle Scholar
This article is published under license to BioMed Central Ltd. This is an open-access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.