Evaluation of reference-based two-color methods for measurement of gene expression ratios using spotted cDNA microarrays
© Peixoto et al; licensee BioMed Central Ltd. 2006
Received: 26 September 2005
Accepted: 24 February 2006
Published: 24 February 2006
Spotted cDNA microarrays generally employ co-hybridization of fluorescently-labeled RNA targets to produce gene expression ratios for subsequent analysis. Direct comparison of two RNA samples in the same microarray provides the highest level of accuracy; however, due to the number of combinatorial pair-wise comparisons, the direct method is impractical for studies including large number of individual samples (e.g., tumor classification studies). For such studies, indirect comparisons using a common reference standard have been the preferred method. Here we evaluated the precision and accuracy of reconstructed ratios from three indirect methods relative to ratios obtained from direct hybridizations, herein considered as the gold-standard.
We performed hybridizations using a fixed amount of Cy3-labeled reference oligonucleotide (RefOligo) against distinct Cy5-labeled targets from prostate, breast and kidney tumor samples. Reconstructed ratios between all tissue pairs were derived from ratios between each tissue sample and RefOligo. Reconstructed ratios were compared to (i) ratios obtained in parallel from direct pair-wise hybridizations of tissue samples, and to (ii) reconstructed ratios derived from hybridization of each tissue against a reference RNA pool (RefPool). To evaluate the effect of the external references, reconstructed ratios were also calculated directly from intensity values of single-channel (One-Color) measurements derived from tissue sample data collected in the RefOligo experiments. We show that the average coefficient of variation of ratios between intra- and inter-slide replicates derived from RefOligo, RefPool and One-Color were similar and 2 to 4-fold higher than ratios obtained in direct hybridizations. Correlation coefficients calculated for all three tissue comparisons were also similar. In addition, the performance of all indirect methods in terms of their robustness to identify genes deemed as differentially expressed based on direct hybridizations, as well as false-positive and false-negative rates, were found to be comparable.
RefOligo produces ratios as precise and accurate as ratios reconstructed from a RNA pool, thus representing a reliable alternative in reference-based hybridization experiments. In addition, One-Color measurements alone can reconstruct expression ratios without loss in precision or accuracy. We conclude that both methods are adequate options in large-scale projects where the amount of a common reference RNA pool is usually restrictive.
Gene expression studies using either oligonucleotides or spotted cDNA microarray platforms are based mainly on data generated in single or dual-channel analysis. While industrially manufactured oligonucleotide arrays (e.g. Agilent Whole Genome 44 k Oligoarray and CodeLink Whole Genome Bioarray) have optimized protocols that perform well on both single and dual-channel microarray experimental designs, custom made spotted cDNA or oligonucleotide microarrays have been mainly employed in two-channel designs . Single-channel (one-color) experiments probe one RNA sample per hybridization, whereas dual-channel (two-color) experiments generate spot signal intensity values from two different RNA samples, each labeled with one of two cyanine dyes (Cy3 or Cy5), followed by simultaneous hybridization. After subtraction of background signal, absolute intensity values, derived from each channel, are often used to calculate expression ratios for subsequent analysis. The ratio of the signal intensities obtained from the two channels is a relative measure of gene expression of the corresponding gene probe. Ratiometric data analysis minimizes various sources of variation related to the construction and hybridization of the microarrays, thus providing the highest level of precision in the comparison of gene expression profiles from two different RNA samples .
While direct hybridization of two experimental RNA samples in the same slide is highly desirable, indirect comparison through the co-hybridization of a test sample together with a common reference standard is the most used experimental design [3, 4]. Relating each experimental sample to a common reference standard facilitates the comparison of ratios across datasets . Several types of reference samples based on commercial Universal reference RNA , genomic DNA [6–8] or PCR products representing the collection of cDNA clones printed on the chip [9, 10] have been proposed, but no single universal reference standard is widely adopted, seriously impeding cross comparisons between different studies. The composition and properties of the selected reference sample must be addressed properly, because it raises issues concerning the experimental design, the goal of the study and the long term comparability of the data. For example, a problem associated to the use of a reference RNA in tumor profiling studies is the requirement of a large amount of high-quality reference sample to allow comparison across multiple datasets [11–13]. Pooling of equal amounts of RNA from test tumor samples is impractical in prospective studies, because samples collected after the construction of the pool would not be represented, precluding adequate comparison of recently collected samples. Cell lines may in principle be an unlimited source of a reference RNA. Indeed several tumor profiling studies have employed such a method [11, 12, 14]. However, the biological variability inherent to cells cultivated in different batches requires that all RNA used to generate the reference pool be prepared prior to the beginning of the hybridizations. As an alternative to reference-based designs, theoretical and experimental work have shown that other types of two-channel designs, namely loop-designs, may produce precise estimates of differential gene expression compared to a design based on a common RNA reference . However, Dobbin and Simon  have demonstrated that for experiments aiming to discover clusters within a collection of samples (class discovery), a common goal of cancer profiling studies, the reference design is more robust than the loop design. According to these authors, variable quality of individual arrays may have a greater impact on cluster analysis when a loop design is used. This consideration is particularly significant in studies using in-house spotted-cDNA microarrays, in which uneven quality between slides of different batches may limit the loop approach.
In this study we designed a set of experiments to evaluate the precision and accuracy of gene expression ratios derived from two-color microarray hybridizations using each of three tumor tissue RNA samples and two different external references: (i) a pooled tumor RNA sample (RefPool) that was labeled in parallel with the test tissue sample; and (ii) a 27-mer reference oligonucleotide (RefOligo) complementary to every feature of the array which was labeled by chemically coupling of a fluorescent nucleotide. The RefOligo method was originally proposed by Dudley et al. to control intensity ratios of gene expression studies in yeast , a system with much lower gene expression complexity. Here, ratios derived from the direct pair-wise hybridizations of human tissue samples were taken as the gold-standard against which ratios derived from reconstructed measurements were compared. These external-reference based ratios were also compared to ratios reconstructed from one-color measurements. The results are discussed based on the strengths and weaknesses of each of the three indirect experimental designs.
Results and discussion
Ratios from direct and indirect hybridizations.
Total number of gene expression ratios calculated from all spots (All) or spots 2 standard deviations (2 SD) above the average slide background are shown. Percentages of ratios above 2 SD are shown in parenthesis.
Kidney vs. Breast
Prostate vs. Breast
Prostate vs. Kidney
Performance among indirect hybridization methods
Correlation coefficient analysis between direct and reconstructed expression ratios.
Only ratios from spots with intensity values 2 SD above the average slide background were used to calculate Pearson's correlations or Spearman's rank-correlations. The number of intensity values used in each correlation analysis is shown (n). All Pearson correlation values were statistically significant (p-value for rejecting Ho: r = 0 << 10-5).
Kidney vs. Breast
Prostate vs. Breast
Prostate vs. Kidney
Direct vs. RefOligo
Direct vs. RefPool
Direct vs. One-Color
Precision of indirect hybridization methods to reconstruct direct measurements
Data presented here show that RefOligo is a reliable alternative to a RNA pool in reference-based hybridization experiments when studying organisms with a complex transcriptome such as humans. The implication of our results is that the unlimited availability of an inexpensive (~US$ 0.30 per hybridization), chemically synthesized reference oligonucleotide makes its use very convenient in large-scale projects where the availability of an RNA pool is usually restrictive. The RefOligo method enables an efficient and flexible experimental design because one may relate expression measurements to a common reference (dual-channel) or, alternatively, use intensity values only (single-channel) . As the RefOligo is complementary to every array element, fluorescent signal derived from bound RefOligo can be used to assess spot quality and facilitates array griddling by image processing software. We speculate that RefOligo will improve comparison of data obtained from different batches of spotted arrays, by correcting for small variations in spot morphology and in the amount of spotted DNA across batches. Moreover, signal intensities of bound reference oligonucleotide molecules correlate well with the amount of cDNA probes present in each spot , suggesting that expression ratios reconstructed from RefOligo may accurately reflect the absolute abundance of each transcript present in the RNA population .
Another important conclusion of our study is that use of One-Color ratios does not compromise precision more than other currently used methods based on indirect measurements. This observation is in line with recent evidence showing that gene classifiers based on intensity measurements may outperform ratio-based classifiers . Intra- and inter-slide expression intensity values obtained with One-Color correlates well and reveal acceptable number of false positives as compared to reference-based methods. Given that direct measurements in datasets containing large number of individual samples (e.g., tumor classification studies) is impractical, One-Color based analysis allows direct comparison of measurements across all samples, with a considerable reduction in costs since it eliminates the requirement of labeling a reference sample
Prostate, kidney and breast tumor samples were obtained from freshly-frozen tissue collections maintained by Instituto Nacional de Câncer, Rio de Janeiro (prostate adenocarcinomas, renal cell carcinomas) and Unidade de Genética e Patologia Moleculares, Hospital do Divino Espírito Santo, Portugal (breast adenocarcinomas). All samples were collected between 1999 and 2002 with informed consent from patients submitted to surgery, and were snap-frozen in liquid nitrogen within 10 min from resection. All samples were examined by a pathologist at each Institution, and, in the case of prostate, hematoxylin/eosin stained micro-sections obtained from each side of the frozen blocks were used to grossly delimit the spatial distribution of the tumor mass. If necessary, tissue blocks were further dissected to warrant that at least 70% of the section used for RNA extraction was composed of malignant cells. Macro-dissected tumor samples were returned to liquid nitrogen until use.
Total RNA was isolated with TRIzol (Invitrogene) using the protocol recommended by the manufacturer. For each tissue, aliquots of RNA isolated from 5 patient samples were pooled, in order to minimize the effect of biological diversity within each set of tissue samples, as well as to circumvent the problem of limited amounts of RNA available from each individual tumor sample. A reference RNA pool (RefPool) was assembled by combining equal amounts of RNA from each one of the three tumor tissues. The amount and quality of each RNA sample was verified on an Agilent 2100 Bioanalyser.
Target labeling and hybridization
RNA samples were labeled with either Cy3- or Cy5-labeled nucleotides (CyScribe first-strand cDNA labeling kit, Amersham Biosciences, Piscataway, NJ), using 10 μg of DNAse-treated RNA and a mixture of oligo-dT and 9-mer random oligonucleotides.
For the indirect hybridizations, Cy5-labeled cDNA targets derived from RNA of each tissue were co-hybridized to either a fixed amount (100 pmol) of a 27-mer 5'-end Cy3-labeled reference oligonucleotide (RefOligo; 5'-CATGATTACGAATTCGAGCTCGGTACC-3', Sigma Aldrich Inc.), or to Cy3-labeled cDNA targets from the reference RNA pool (RefPool) labeled as described above with CyScribe first-strand cDNA labeling kit (Amersham Biosciences, Piscataway, NJ). The 27-mer RefOligo has no sequence similarity to any human expressed sequence available in GenBank. For RefOligo and RefPool, two replicate hybridizations were performed for each tissue. For direct hybridizations, fluorescently-labeled targets derived from each test tissue sample were combined pair-wise. A replicate hybridization with dye-reversal was performed for each pair of tissues.
Targets were evaporated in a Speedvac, resuspended in hybridization solution (50% formamide, 25% Amersham Microarray Hybridization Buffer V.2, 25% H2O) and manually hybridized for 16 h at 42°C to microarray slides containing 4,608 different cDNA probes, each spotted in duplicate on either half of the slide. A detailed description of the spotted cDNA microarray platform used is presented elsewhere . Following hybridization, slides were washed (1.0× SSC, 0.2% SDS 10 min. at 55°C, 0.1× SSC, 0.2% SDS 10 min. at 55°C, 0.1× SSC, 0.2% SDS 10 min. at 55°C, 0.1× SSC 1 min. at RT, 0.1× SSC 1 min at RT, dH2O 10 sec. at RT) and dried with a N2 stream. Processed slides were scanned with a PMT setting of 700 V (GenIII Scanner – Amersham Biosciences) and background-subtracted artifact-removed median intensities of both Cy3 and Cy5 emissions were extracted for each spot from raw images using ArrayVision V.7.2 software (Imaging Research Inc., Ontario, Canada).
Data normalization, filtering and averaging of replicates
To correct for systematic biases on the data originated from small differences in the labeling and/or detection efficiencies between the fluorescent dyes, both direct and reconstructed expression ratios were logged (base 2) and normalized using a locally weighted linear regression (LOWESS) algorithm [2, 26] implemented as scripts written in R language . Unless indicated, only spots whose background-subtracted intensities measured in both channels were 2 standard deviations above the local background (defined for each sub-array by a set of plant cDNA negative control probes) were considered in the analysis. For indirect ratio reconstructions, LOWESS normalization was performed in the M vs. A space, where:
A = log2 [test sample 1(cy5)]/2 + log2 [test sample 2(cy5)]/2
As there were 2 technical replicates for each indirect hybridization, and each cDNA probe was deposited in duplicate in each slide, 8 possible reconstructed expression ratio values could be generated for a given cDNA probe: K1L/B1L, K1R/B1R, K1L/B2L, K1R/B2R, K2L/B1L, K2R/B1R, K2L/B2L and K2R/B2R, where K and B denote Kidney and Breast for example, 1 and 2 denote different hybridizations and R and L denote the right and left spot sets from each slide, respectively. Final ratio values was obtained by taking the median value of all 8 reconstructed ratios for each cDNA probe, for RefOligo, RefPool and One-Color comparisons.
For direct hybridizations, LOWESS normalization was performed by combining log2 ratios from dye-swap replicate experiments as described in . For direct hybridizations, LOWESS normalization was performed in the M vs. A space where:
Mdir = 0.5 * log2(cy5/cy3 * cy3'/cy5') =
= 0.5 * log2(sample 1 (cy5)/sample2(cy3) * sample 1(cy3')/sample 2(cy5'))
A = 0.25 * log2(Samaple 1(cy3) * sample 1(cy5) * sample 2(cy3) * sample 2(cy5))
Raw and processed data files from direct and indirect hybridizations are available at author's website .
Precision of intra-slide replicates from direct or reconstructed ratios was estimated by calculating, for each tissue comparison, the average coefficient of variance (CV) of the two replicated spots representing the same cDNA that are present in the microarrays. Final average intra-slide CV +/- SD from RefOligo or RefPool was calculated from the average intra-slide CV of each method measured in each tissue comparison. Average inter-slide CV was estimated by calculating the average CV of reconstructed ratios measured across the same spots in different slides.
To evaluate the variance of expression log2-ratios reconstructed from RefOligo or RefPool along the range of expression intensities, MA-plots were generated for each hybridization method (for each tissue comparison) using M and A values calculated from all replicates as defined above. For this analysis, the 2 SD intensity filter was not applied to access the variance on the entire range of intensities. Next, a sliding window of 1.5, moving 0.4 at each step, was used to calculate the variance of reconstructed ratios M along the range of expression intensities A (X axis). These analyses were performed using scripts written in R . The R scripts used in all analyses described in the present work are available at author's website .
To evaluate the accuracy of expression ratios reconstructed from RefOligo, RefPool and One-Color measurements, a set of genes differentially expressed between each pair of tissues based on the direct hybridizations was identified, and was defined as the gold-standard set. Genes that were differentially expressed in the pair-wise tissue comparisons were selected with the statistical approach described in . In short, the HTself method classifies an expression ratio as significant or not according to an experimentally derived intensity-dependent fold-change cutoff . These cutoffs are obtained from experiments where fluorescent targets derived from the same RNA and labeled with either Cy3 or Cy5 are co-hybridized to the same microarray. For direct comparisons we performed two self-self direct hybridizations for each tissue. Thus, 6 replicate ratios for each spotted probe were generated: K1L/K2L, K1R/K2R, P1L/P2L, P1R/P2R, B1L/B2L, B1R/B2R, where K, P and B denote Kidney, Prostate and Breast RNAs, 1 and 2 denote either Cy3 or Cy5 dye labeling, and R and L denote the right and left array sets from each slide, respectively. This procedure yielded the experimental null distribution of the differential expression significance test since, by definition, there is no differential expression in the self-self dataset (see details in ). Concerning the accuracy of the gold-standard, it should be noted that this set of genes was selected by applying a statistical approach based on self-self hybridizations using a single RNA labeled with both Cy3 and Cy5. Therefore, any compression on the ratios due to systematic dye effects would also be present in the ratio cut-offs derived from these self-self experiments, thus cancelling out most of the possible bias in selecting the gold-standard set of genes. To derive a null distribution for the indirect comparisons, pseudo self-self ratios were reconstructed from replicate RefOligo, RefPool and One-Color data derived from the same RNA obtained from separate arrays. A total of 6 self-self ratios were thus generated for each RefOligo, RefPool and One-Color dataset. Next, we created 95% credibility intervals for both direct and indirect self-self log2-ratios [see Additional file 3]. These intervals were used to classify a given gene as differentially expressed if its replicate ratios in the pair-wise tissue comparisons were consistently (> 50%) outside the credibility interval thresholds. To be stringent we considered in this test only genes with more than 4 valid reconstructed ratios for the indirect dataset and only genes with all valid ratios for the direct dataset. Concordance and discordance among gene sets identified by direct and reconstructed ratios were represented as Venn diagrams.
The authors are grateful to Drs. Vitor Carneiro, Victor Santos, Teresa Eloi and Laura de Fez Sayas from Hospital do Divino Espírito Santo, Portugal, and to Drs. Marcello Barcinski and Franz Campos from Instituto Nacional de Câncer, Rio de Janeiro for providing tumor tissues and RNA samples. We thank Dr. Ricardo DeMarco for valuable suggestions and critically reading the manuscript. B.R.P. is recipient of a postdoctoral fellowship from Science and Technology Foundation, Portugal. This work was mostly supported by a grant from Fundação de Amparo a Pesquisa do Estado de São Paulo, Brasil (FAPESP) to S.V.A. Additional resources were provided by a grant from Luso-American Foundation to B.R.P (Ref. L-V-383/2002); and fellowships from Conselho Nacional de Desenvolvimento Científico e Tecnológico (CNPq) and FAPESP.
- Churchill GA: Fundamentals of experimental design for cDNA microarrays. Nat Genet. 2002, 32 Suppl: 490-495. 10.1038/ng1031.PubMedView ArticleGoogle Scholar
- Quackenbush J: Microarray data normalization and transformation. Nat Genet. 2002, 32 Suppl: 496-501. 10.1038/ng1032.PubMedView ArticleGoogle Scholar
- Kerr MK, Churchill GA: Experimental design for gene expression microarrays. Biostatistics. 2001, 2: 183-201. 10.1093/biostatistics/2.2.183.PubMedView ArticleGoogle Scholar
- Yang YH, Speed T: Design issues for cDNA microarray experiments. Nat Rev Genet. 2002, 3: 579-588.PubMedGoogle Scholar
- Neben K, Korshunov A, Benner A, Wrobel G, Hahn M, Kokocinski F, Golanov A, Joos S, Lichter P: Microarray-based screening for molecular markers in medulloblastoma revealed STK15 as independent predictor for survival. Cancer Res. 2004, 64: 3103-3111. 10.1158/0008-5472.CAN-03-3968.PubMedView ArticleGoogle Scholar
- Kim H, Zhao B, Snesrud EC, Haas BJ, Town CD, Quackenbush J: Use of RNA and genomic DNA references for inferred comparisons in DNA microarray analyses. Biotechniques. 2002, 33: 924-930.PubMedGoogle Scholar
- Williams BA, Gwirtz RM, Wold BJ: Genomic DNA as a cohybridization standard for mammalian microarray measurements. Nucleic Acids Res. 2004, 32: e81-10.1093/nar/gnh078.PubMedPubMed CentralView ArticleGoogle Scholar
- Gadgil M, Lian W, Gadgil C, Kapur V, Hu WS: An analysis of the use of genomic DNA as a universal reference in two channel DNA microarrays. BMC Genomics. 2005, 6: 66-10.1186/1471-2164-6-66.PubMedPubMed CentralView ArticleGoogle Scholar
- Gorreta F, Barzaghi D, VanMeter AJ, Chandhoke V, Del Giacco L: Development of a new reference standard for microarray experiments. Biotechniques. 2004, 36: 1002-1009.PubMedGoogle Scholar
- Sterrenburg E, Turk R, Boer JM, van Ommen GB, den Dunnen JT: A common reference for cDNA microarray hybridizations. Nucleic Acids Res. 2002, 30: e116-10.1093/nar/gnf115.PubMedPubMed CentralView ArticleGoogle Scholar
- Alizadeh AA, Eisen MB, Davis RE, Ma C, Lossos IS, Rosenwald A, Boldrick JC, Sabet H, Tran T, Yu X, Powell JI, Yang L, Marti GE, Moore T, Hudson JJ, Lu L, Lewis DB, Tibshirani R, Sherlock G, Chan WC, Greiner TC, Weisenburger DD, Armitage JO, Warnke R, Levy R, Wilson W, Grever MR, Byrd JC, Botstein D, Brown PO, Staudt LM: Distinct types of diffuse large B-cell lymphoma identified by gene expression profiling. Nature. 2000, 403: 503-511. 10.1038/35000501.PubMedView ArticleGoogle Scholar
- Perou CM, Sorlie T, Eisen MB, van de Rijn M, Jeffrey SS, Rees CA, Pollack JR, Ross DT, Johnsen H, Akslen LA, Fluge O, Pergamenschikov A, Williams C, Zhu SX, Lonning PE, Borresen-Dale AL, Brown PO, Botstein D: Molecular portraits of human breast tumours. Nature. 2000, 406: 747-752. 10.1038/35021093.PubMedView ArticleGoogle Scholar
- van 't Veer LJ, Dai H, van de Vijver MJ, He YD, Hart AA, Mao M, Peterse HL, van der Kooy K, Marton MJ, Witteveen AT, Schreiber GJ, Kerkhoven RM, Roberts C, Linsley PS, Bernards R, Friend SH: Gene expression profiling predicts clinical outcome of breast cancer. Nature. 2002, 415: 530-536. 10.1038/415530a.PubMedView ArticleGoogle Scholar
- Garber ME, Troyanskaya OG, Schluens K, Petersen S, Thaesler Z, Pacyna-Gengelbach M, van de Rijn M, Rosen GD, Perou CM, Whyte RI, Altman RB, Brown PO, Botstein D, Petersen I: Diversity of gene expression in adenocarcinoma of the lung. Proc Natl Acad Sci U S A. 2001, 98: 13784-13789. 10.1073/pnas.241500798.PubMedPubMed CentralView ArticleGoogle Scholar
- Vinciotti V, Khanin R, D'Alimonte D, Liu X, Cattini N, Hotchkiss G, Bucca G, de Jesus O, Rasaiyaah J, Smith CP, Kellam P, Wit E: An experimental evaluation of a loop versus a reference design for two-channel microarrays. Bioinformatics. 2005, 21: 492-501. 10.1093/bioinformatics/bti022.PubMedView ArticleGoogle Scholar
- Dobbin K, Simon R: Comparison of microarray designs for class comparison and class discovery. Bioinformatics. 2002, 18: 1438-1445. 10.1093/bioinformatics/18.11.1438.PubMedView ArticleGoogle Scholar
- Dudley AM, Aach J, Steffen MA, Church GM: Measuring absolute expression with microarrays with a calibrated reference sample and an extended signal intensity range. Proc Natl Acad Sci USA. 2002, 99: 7554-7559. 10.1073/pnas.112683499.PubMedPubMed CentralView ArticleGoogle Scholar
- Ramaswamy S, Ross KN, Lander ES, Golub TR: A molecular signature of metastasis in primary solid tumors. Nat Genet. 2003, 33: 49-54. 10.1038/ng1060.PubMedView ArticleGoogle Scholar
- Petersen D, Chandramouli GV, Geoghegan J, Hilburn J, Paarlberg J, Kim CH, Munroe D, Gangi L, Han J, Puri R, Staudt L, Weinstein J, Barrett JC, Green J, Kawasaki ES: Three microarray platforms: an analysis of their concordance in profiling gene expression. BMC Genomics. 2005, 6: 63-10.1186/1471-2164-6-63.PubMedPubMed CentralView ArticleGoogle Scholar
- Glinsky GV, Berezovska O, Glinskii AB: Microarray analysis identifies a death-from-cancer signature predicting therapy failure in patients with multiple types of cancer. J Clin Invest. 2005, 115: 1503-1521. 10.1172/JCI23412.PubMedPubMed CentralView ArticleGoogle Scholar
- t Hoen PA, Turk R, Boer JM, Sterrenburg E, de Menezes RX, van Ommen GJ, den Dunnen JT: Intensity-based analysis of two-colour microarrays enables efficient and flexible hybridization designs. Nucleic Acids Res. 2004, 32: e41-10.1093/nar/gnh038.View ArticleGoogle Scholar
- Vencio RZ, Koide T: HTself: Self-Self Based Statistical Test for Low Replication Microarray Studies. DNA Res. 2005, 12: 211-214.PubMedView ArticleGoogle Scholar
- Rouse RJ, Espinoza CR, Niedner RH, Hardiman G: Development of a microarray assay that measures hybridization stoichiometry in moles. Biotechniques. 2004, 36: 464-470.PubMedGoogle Scholar
- Attoor S, Dougherty ER, Chen Y, Bittner ML, Trent JM: Which is better for cDNA-microarray-based classification: ratios or direct intensities. Bioinformatics. 2004, 20: 2513-2520. 10.1093/bioinformatics/bth272.PubMedView ArticleGoogle Scholar
- Reis EM, Nakaya HI, Louro R, Canavez FC, Flatschart AV, Almeida GT, Egidio CM, Paquola AC, Machado AA, Festa F, Yamamoto D, Alvarenga R, da Silva CC, Brito GC, Simon SD, Moreira-Filho CA, Leite KR, Camara-Lopes LH, Campos FS, Gimba E, Vignal GM, El-Dorry H, Sogayar MC, Barcinski MA, da Silva AM, Verjovski-Almeida S: Antisense intronic non-coding RNA levels correlate to the degree of tumor differentiation in prostate cancer. Oncogene. 2004, 23: 6684-6692. 10.1038/sj.onc.1207880.PubMedView ArticleGoogle 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: e15-10.1093/nar/30.4.e15.PubMedPubMed CentralView ArticleGoogle Scholar
- Koide T, Zaini PA, Moreira LM, Vencio RZ, Matsukuma AY, Durham AM, Teixeira DC, El-Dorry H, Monteiro PB, da Silva AC, Verjovski-Almeida S, da Silva AM, Gomes SL: DNA microarray-based genome comparison of a pathogenic and a nonpathogenic strain of Xylella fastidiosa delineates genes important for bacterial virulence. J Bacteriol. 2004, 186: 5442-5449. 10.1128/JB.186.16.5442-5449.2004.PubMedPubMed CentralView ArticleGoogle Scholar
- Supplementary information at author's website. [http://verjo19.iq.usp.br/gec/en/publications/Peixoto_et_al/]
- The R Project for Statistical Computing. [http://www.r-project.org/]
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.