| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Molecular Diagnostics and Genetics |
Departments of1 Pathology,2 Biostatistics, and3 Surgery, Virginia Commonwealth University, Richmond, VA.
aAddress correspondence to this author at: Molecular Diagnostics Division, Department of Pathology, Virginia Commonwealth University, Clinical Support Center Building, Room 247, 403 North 13th St., PO Box 980248, Richmond, VA 23298-0248. Fax 804-225-4738; e-mail agonzalez{at}hsc.vcu.edu.
 |
Abstract |
|---|
 Top Abstract IntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
Methods: In the present studies we developed quality-control criteria and evaluated their effect in microarray data analysis using total RNA from cell lines, frozen tumors, and a commercially available reference RNA. Quality-control criteria such as A260/A280 ratios, percentage of rRNA, and median size of cDNA and cRNA synthesis products were evaluated for robustness in microarray analysis. Furthermore, precision studies using a reference material were performed on the Affymetrix® HG-U133A high-density oligonucleotide microarrays. The same reference RNA sample was examined in 16 different chips run on 2 different days in the four different modules of the Affymetrix fluidics workstation. Fresh and frozen fragmented cRNAs were also compared. An ANOVA model was fit to identify the main sources of variation.
Results: Good-quality samples showed >30% rRNA in the electropherograms and cDNA and cRNA synthesis products with median sizes of 2.0 and 3.0 kb, respectively. Precision studies showed that the main source of variation was the day-to-day variability, minimally affecting hybridization exogenous control genes. Altogether, the results showed that the Affymetrix Genechip® system is highly reproducible when RNA that meet the quality-control criteria are used (overall P >0.01).
Conclusions: These results confirm the need to establish defined quality-control criteria for sample quality to distinguish between analytical and biological variability.
|
Introduction |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
The issue of quality control (QC)1 and variability of data generated from microarray technology is recognized as critical to the potential use of microarrays in clinical applications. Assessing the quality of the initial RNA sample is as critical as ensuring successful cDNA and cRNA synthesis before microarray hybridization. However, QC criteria for total RNA samples and subsequent synthesis products used in microarray analysis have yet to be defined.
Monitoring the quality of the initial RNA sample is as important as choosing the appropriate statistical analysis because the quality of the sample might strongly influence the diagnostic and predictive power obtained from the microarray data. It is anticipated that application of different algorithms to reduce the number of genes to a nonredundant informative set would yield a set of genes useful for clustering tumor specimens into groups that share common features(3). Previous reports have shown that expression of known tumor markers or expression patterns correlating with the response to a specific therapy can create characteristic profiles of the groups. These profiles can then be used to identify newly acquired unknown specimens(4)(5). Nevertheless, a class prediction algorithm applied to an unknown sample of poor quality may lead to misclassification. A recent study showed that data quality was decreased when moderately degraded RNA samples were used(6). Furthermore, severely degraded RNA samples demonstrated low overall quality and poor correlation with their intact sample counterparts. The authors concluded that such samples should be used with caution if considered for microarray analysis and subsequent classification(6).
The standard Affymetrix protocol uses as starting material 5–40 µg of high-quality total RNA, which is used as a template for the synthesis of cDNA molecules from which biotinylated cRNA is synthesized. All procedures associated with probe preparation, hybridization, washing, and scanning have been standardized(7). Because microarray technologies in general offer a complex association between expression of a particular gene and the ultimate light intensity measured by the scanner, the accuracy and consistency of the data need to be ensured. Multiple sources of technical variation can contribute to the final observed gene expression obtained with microarrays. Some of the sources of technical variation could be different lot numbers for reagents and/or chips, different individuals handling the whole or part of the process, and day-to-day variability. In the Affymetrix platform, each sample is hybridized and stained independently in a single chip per module, adding the module-to-module effect to the list of different sources of variation. In addition, the biological sample loaded in every chip is composed of chemically fragmented (frg) cRNA molecules that can be freshly prepared immediately before hybridization or can be stored frozen until needed. The freezing process may be considered another source of variability that may need to be taken into account in the search for relevant biological discoveries. Historically, distinguishing between significant and nonsignificant biological changes in gene expression depended on the number of replicates analyzed and on the fold-change cutoff. This cutoff varied from 2- to 2.5-fold depending on the researcher’s criteria and experimental design(8)(9)(10). Such strategies for microarray data analysis do not account for the different sources of variation introduced by the system.
In this study we focused our attention on a subset of the above-mentioned sources of variation to address the day-to-day, module-to-module, and fresh/frozen frg cRNA effects on the same biological sample. We used 16 replicate GeneChips to study these components of variability. We also evaluated the quality of gene expression data, using measurable QC criteria that take into account the quality of the total RNA samples that were isolated from different sample types. In addition, cDNA and cRNA synthesis products were monitored to account for technical variability during sample preparation that could negatively impact the quality of microarray data.
|
Materials and Methods |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
affymetrix genechip standard protocol
The Affymetrix standard protocol has been described extensively elsewhere(7)(13). Briefly, starting with 5 µg of total RNA from every sample, we generated double-stranded cDNA using a 24mer oligodeoxythymidylic acid primer with a T7 RNA polymerase promoter site added to the 3' end (Superscript cDNA Synthesis System; Life Technologies, Inc.). After second-strand synthesis, in vitro transcription was performed with the Enzo BioArray High Yield RNA Transcript Labeling Kit (Enzo Diagnostics) to produce biotin-labeled cRNA. We fragmented 20 µg of the cRNA product and hybridized it for 18–20 h into HG-133A microarrays containing 22 283 probe sets. Each microarray was washed and stained with streptavidin-phycoerythrin and was scanned at a 6-µm resolution by the Agilent G2500A Technologies Gene Array scanner (Agilent Technologies) according to the GeneChip Expression Analysis Technical Manual procedures (Affymetrix). After scanning of the chips, the raw intensities for every probe were stored in electronic files (in .DAT and .CEL formats) by the Microarray Suite 5.0 software (Affymetrix). The .CEL files from this study can be accessed through the following website: http://www.ctrf-cagenomics.vcu.edu/publiclyavaildata.htm (This website can be viewed only with a browser such as Microsoft® Internet Explorer, Ver. 6.0 or greater).
hybridization qc criteria
Immediately after scanning of every chip, several variables were monitored for QC purposes: the scaling factor; the noise (RawQ)(14) value for a given probe array hybridization; the percentage of probe sets declared "Present" (%P) by the detection call algorithm; the 3'/5' ratios of the signal intensity values for two housekeeping genes, glyceraldehyde-3-phosphate dehydrogenase (GAPDH) and ß-actin; and the presence or absence call for the ribosomal RNAs (rRNAs) 18S and 28S. The scaling factor is used to scale all probe sets to an arbitrary target value (TGT); in this study, the TGT value was set at 100. The RawQ value is calculated by taking the mean, over all the cells used in background computation, of the following value for each cell: standard deviation of the pixel intensity divided by the square root of the pixel number.
study design to assess precision
The precision study experiment was designed to assess variation attributable to day, use of fresh/frozen cRNA, and the four different modules of the Affymetrix fluidics workstation. To eliminate operator variations, the same person completed the synthesis and hybridization of all 16 chips run for this study. Five identical aliquots of 40 µg of the Reference RNA were used as template for cDNA synthesis, cRNA in vitro transcription, and labeling reactions. The final frg cRNA aliquots were pooled together. In addition, on the first day, four chips were run with fresh frg cRNA and four chips were run with frozen frg cRNA, using the same hybridization solution. On the second day, a fresh hybridization solution was prepared to replicate the entire experiment with frozen frg cRNA (see Table 1
for the complete design). Hybridization was performed using the same lot number of the HG-U133A arrays to minimize chip-to-chip variability.
|
expression values
For every probe set, normalization, background subtraction, and expression summaries were calculated by three commonly used methods. First the Microarray Suite 5.0 (MAS5) method (Affymetrix) was used to obtain probe set summaries. A detailed description of this algorithm has been published elsewhere(15). Numerical expression summaries were stored in electronic files (in .CHP format). Next, model-based expression indexes (MBEI)(16) were calculated by use of a multiplicative model to account for probe affinity effects in calculating probe set expression summaries. Finally, the robust multiarray average (RMA) method was used(17); this method uses quantile normalization followed by a median polish to remove probe affinity effects when calculating probe set summaries.
statistical analysis
We used ANOVA on each probe set to analyze the significance of each separate identifiable sources of variation (i.e., day, module, frozen, and the day x module and frozen x module two-way interactions) on gene expression. In addition, the significance of any of the above effects was assessed by the whole model test (with 11 degrees of freedom). The ANOVA model was fit to the MAS5, MBEI, and RMA probe set summaries separately. Additionally, we used different statistical methods that are widely used by the microarray community for class comparison to identify genes differentially expressed among the experimental conditions in our study. First, for each probe set an F-test was performed. In addition, univariate permutation-based P values, using 2000 permutations, were calculated for the same class comparisons. Both the F-test and the permutation-based P values were obtained with use of BRB-ArrayTools, Ver. 3.1.0(18), an Excel Add-in that performs analyses of microarray data. The significance analysis of microarrays (SAM)(19) method was also used, which allows one to control for the false discovery rate. The publicly available SAM Excel Add-in, developed at Stanford University (Stanford, CA) was used to conduct this analysis. To account for multiple comparisons, we set the 
to 0.0001 for declaring probe sets as significantly differentially expressed, implying that of the 22 283 probe sets present in the HG-U133A array, we expected by chance that 
2 probe sets on the list will be false positives.
|
Results |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
1.8 and A260/A270 1.1 when the RNA preparation was free from protein and/or phenol contamination, respectively. In addition, the two most distinct and intense peaks on the electropherogram obtained with the bioanalyzer, corresponding to the 18S and 28S rRNAs, should show 28S/18S ratios >1.4, with the sum of both rRNA peak areas accounting for more than 30% of all RNA, indicating that the RNA sample was intact or undegraded (Table 2
). We also included as QC criteria for the sample processing, the median cDNA and cRNA synthesis product length, where medians of 2.0 and 3.0 kb were considered optimal for cDNA and cRNA synthesis, respectively. Finally, total RNA samples meeting all of the above-mentioned QC criteria typically showed ratios of the signal intensities for probe sets designed to match the 3' and 5' regions of two housekeeping transcripts, GAPDH and ß-actin, near 1.00 (Table 2
).
|
RNA samples from different sources were tested for QC purposes. Good-quality RNA was obtained from the Stratagene Universal Human Reference RNA sample. It met almost all of the QC criteria except for the 28S/18S rRNA ratio, which was <1.4 (Table 2
). Nevertheless, the percentage of rRNA was >30%, and the cDNA and cRNA syntheses were successful as indicated by median lengths of 2.0 and 3.0 kb for the respective products.
In addition, we obtained good-quality RNA from freshly harvested cell lines, using the TRIZOL procedure. These samples systematically met all QC criteria (Table 2
and Fig. 1A
). Interestingly, total RNA isolated from snap-frozen tissue samples by the same procedure did not meet the QC criteria, leading to poor-quality cDNA and cRNA products (Fig. 1B
). The RNA isolated from these samples showed good A260/A280 (>1.8) and A260/A270 ratios (>1.1). In addition, these samples were intact or undegraded, with rRNA 28S/18S ratios >1.4 and percentages of rRNA >30%. However, as clearly shown in Table 2
, they systematically failed to produce cDNA and, therefore, cRNA molecules >300 and 700 bp, respectively. Moreover, all of these samples showed GAPDH and ß-actin 3'/5' ratios much greater than 1.00, which was in agreement with unsuccessful cDNA and cRNA synthesis (Table 2
). Suspecting the presence of inhibitors in these RNA preparations, which might be inherent to the tissue-handling process, we reextracted RNA from the same snap-frozen tissue samples with the TRIZOL protocol and followed with an extra RNeasy cleanup step. We found that RNA samples extracted with this modified procedure met all of the QC criteria, with cDNA and cRNA synthesis products within median lengths of 2.0 and 3.0 kb, respectively (Fig. 1C
) and 3'/5' ratios close to 1.00 for both housekeeping genes, GAPDH and ß-actin (Table 2
).
|
On the other hand, of 27 frozen bone marrow samples, 5 showed different patterns of RNA degradation attributable to inadequate handling and storage conditions. These five samples failed to meet the QC criteria, showing electropherogram profiles consistent with RNA degradation. A typical result is shown in Fig. 1D
. Accordingly, these samples showed low percentages of rRNA and shorter cDNA and cRNA products. The results from two representative bone marrow samples exhibiting different degrees of degradation are summarized in Table 2
(for complete results, see Table S1 in the Data Supplement that accompanies the online version of this article at http://www.clinchem.org/content/vol50/issue11/). Interestingly, poor-quality RNA was obtained from snap-frozen tumor samples containing different degrees of necrosis. The results from a representative tissue specimen containing 80% necrosis are summarized in the last row of Table 2
. As expected, the RNA failed to meet the RNA QC criteria, with a percentage of rRNA <30%. This reflected the presence of degraded or partially degraded RNA (Fig. 1E
). Although the overall yield of the cDNA and cRNA syntheses were notably less than the usual yield, the median lengths of the cDNA and cRNA products were 2.0 and 3.0 kb, respectively (Table 2
). This suggested that the RNA sample contained a subset of intact RNA, likely arising from the viable tumor cells located in the remaining 20% of the tissue sample.
Overall, our results strongly suggest that the rRNA content calculated from the electropherogram profile, in conjunction with the size distribution of the cDNA and cRNA products, were good indicators of the integrity and purity of total RNA samples, respectively.
precision study
We analyzed the significance of the changes observed in different chip characteristics attributable to day-to-day, module-to-module, and fresh/frozen frg cRNA sources of variations, as well as the interactions day x module and frozen x module. In general, the different characteristics analyzed for each HG-U133A chip showed almost no significant change across the 16 chips (P >0.05; Table 3
). However, the RawQ value was significantly affected by the day effect (P <0.05), and the probe set designed to bind to the 3' end of the ß-actin transcript was significantly affected by all the tested sources of variation (P <0.005), as indicated in bold font in Table 3
.
|
The 22 283 probe sets arrayed on the HG-U133A chip were first analyzed with the MAS5 probe set expression summaries. The response variable was the log2 of the signal intensity. After fitting of the model, the log2 estimate of the least-squares mean was used to assess the fold change and the significance of that change for each effect included in this study by probe set. Panels A–C in Fig. 2
display the fresh/frozen, day-to-day, and module-to-module estimates, respectively, plotted against negative log10 P value when the ANOVA included all probe sets; panels D–F display the model results when the ANOVA was restricted to Present probe sets. Fresh/frozen frg cRNA was not a significant source of variation (Fig. 2
, A and D) in either all-probe sets or in Present probe sets. On the other hand, day-to-day was a significant source of variation (P <0.0001), affecting the gene expression results for at least two Present probe sets (Fig. 2E
). One of these probe sets queries a hybridization exogenous control gene. The largest dispersion of the data was observed for the module-to-module effect. However, most of the changes were not significant at a P value <0.0001 (Fig. 2
, C and F). Overall, the ANOVA restricted to Present probe sets demonstrated a tight dispersion, below a twofold change cutoff. Nevertheless, we observed up to eightfold changes among some "Absent" probe sets and among those with low signal intensities. To further investigate the impact of Absent probe sets, we summarized the results from the ANOVA model depending on how many of the n = 16 observations were labeled as Present or Absent by the MAS5 detection call algorithm. For 98.84% of the probes (22 023 of 22 283), there were no significant effects in the model (overall P >0.01). However, 15 probe sets called Absent and 8 probe sets called Present showed significant effects in the ANOVA model (overall P <0.001; Table 4
). Similar results were obtained when we performed the ANOVA on the expression summaries calculated with MBEI and RMA. As expected with highly robust methods of analysis, the ANOVA results issued from the RMA algorithm showed less dispersion when we analyzed all probe sets compared with the values calculated with MAS5 (data not shown).
|
|
Because all 16 chips were run with separate aliquots from the same RNA source, any differences would reflect differences in sample handling and labeling or from random variation. We used the following algorithms to calculate probe set expression summaries: MAS5, MBEI, and RMA.
From the ANOVA results represented by the "volcano" plots in Fig. 2
, we found that the day-to-day effect was the only source of technical variability that significantly affected the expression values of two Present probe sets. Hence, we decided to investigate the impact of that effect on gene expression by performing class comparisons between days 1 and 2, using SAM and permutation-based tests. Expression summaries for 10 probe sets were declared significantly different by both methods for the three different expression summaries (Fig. 3
). These 10 probe sets had significantly higher intensities on day 1 than on day 2. Interestingly, these were probe sets designed to detect the set of hybridization exogenous controls recommended by Affymetrix: BioB, BioC, BioD, and Cre(7).
|
|
Discussion |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
2.0(20), in good-quality total RNA samples, we have seen successful cDNA and cRNA syntheses with 28S/18S rRNA ratios as low as 0.8 (Table 2
). Furthermore, we have tested RNA samples isolated from mishandled bone marrow specimens that displayed different degrees of degradation, and the 28S/18S rRNA ratios were >1.5 (see Table S1 in the online Data Supplement). We therefore suggest that the percentage of rRNA in the electropherograms might be a suitable QC criterion to evaluate total RNA integrity instead of just the 28S/18S rRNA ratio. These results are in agreement with previously reported data, where it was shown that the 28S/18S ratio for a particular RNA sample has no practical value and should be used only along with the absence of prominent degradation products(6). Here we demonstrated that the sole use of the 28S/18S ratio as a QC criterion could be misleading. More importantly, when examining the total RNA electropherogram, one must also consider the possible sources of degradation. Our results showed that RNA degradation caused by sample mishandling can vary depending on the extent of technical damage, with the rRNA contribution ranging from 0% in completely degraded RNA samples to 20% in partially degraded RNA samples. From the results obtained in this study, we established a cutoff value of 30% rRNA contribution to the total area under the electropherogram for RNA samples to be considered as intact or undegraded (see Figs. S1 and S2 in the online Data Supplement). On the other hand, tissue samples containing different degrees of necrosis yielded a mixture of intact and degraded RNA, producing electropherograms with <30% rRNA contribution and variable 28S/18S ratios. In these samples, however, the cDNA and cRNA syntheses were performed successfully, as opposed to those that were performed with RNA samples that were degraded by different technical causes.
Our results with freshly harvested cell lines showed that high-quality RNA could be obtained from a simple TRIZOL extraction; however, snap-frozen tissues required a subsequent RNA cleanup process to successfully synthesize cDNA and cRNA molecules. As a standard practice, it would be advisable to evaluate every different type of sample with all of the QC criteria defined in this study to ensure consistency in sample handling and processing. Otherwise, microarray analysis may misclassify technical variability as biological variability.
The performance of every chip was also evaluated for indications of successful hybridization, washing, and staining procedures. A set of control genes within every chip can be used to monitor the sample quality at the hybridization step. Thus, 3'/5' ratios close to 1.0 for housekeeping genes such as GAPDH and ß-actin are expected values for good-quality samples and efficient extension reactions during the cDNA and cRNA synthesis steps. Moreover, values such as the noise (RawQ or the pixel to pixel variation for very low intensity cells within the chip), scaling factor, and percentage of Present probe sets (%P) are very good indicators of outlier samples or chips within similar sample types and runs(21). From the results of the present study, we concluded that the %P was a good indicator of quality of the microarray, correlating with the quality of the sample only when comparisons were made within the same tissue or cell type. Thus, when we compared two different RNA extraction protocols applied on five different ovarian tumor samples, a significant difference (paired t-test, P <0.01) was observed in the %P calls. Hence, ovarian tumor RNA extracted with the TRIZOL procedure showed a mean (SD) %P calls of 52.00 (3.60), whereas for the same ovarian tumor RNA extracted with the TRIZOL procedure followed by a QIAGEN cleanup step, the %P calls were significantly increased, to 57.66 (1.48).
Among all hybridization criteria analyzed in the present study, we found that only the RawQ value and the expression summary of the 3' probe set querying ß-actin were significantly (P <0.05) affected by some of the sources of variation analyzed. The significant day-to-day effect on the RawQ value was probably attributable to the fact that the hybridization and staining solutions were freshly prepared every day. Thus, slight differences in the final composition of the above-mentioned solutions could impact the overall noise (RawQ value) of individual chips run on different days. Surprisingly, the expression summary of the 3' probe set querying ß-actin seemed to be significantly affected by every effect analyzed; however, the extent of the changes was <1.2-fold for all sources of variation (data not shown). Moreover, the fact that the 5' probe set querying the same transcript was not significantly affected by any of the effects suggests that the changes observed in the 3' probe set might be related to a rather less robust probe design than actual differences in the concentration of that particular transcript.
In this study, we used an ANOVA model to find significant changes in the signal intensity values attributable to all identifiable technical effects. Using the MAS5 algorithms, we found a subset of 15 significantly altered probe sets (P <0.001), corresponding to Absent probe sets, with signal intensities below the mean (SD) background for all chips [50.70 (2.21)]. We therefore concluded that the variation of the values for these probe sets should be considered to be related to noise, or random variation, more than to any of the sources of variation analyzed in this study.
On the other hand, we found a subset of eight significantly affected probe sets (P <0.001) corresponding to Present probe sets according to the MAS 5.0 algorithms. Seven of them had signal intensities <200, and the variation across the 16 chips was always <1.25-fold change. The eighth probe set from this group corresponded to the Filamin 1 gene, with a mean signal intensity value of 610. Because that value was nearly 10 times the background value for this chip, we considered this probe set a good candidate for false discovery because of its large variability across the technique itself, not reflecting the biological differences of these 16 replicates. To evaluate the performance of this particular probe set, we analyzed another probe set that interrogates the same gene, Filamin 1, designed to hybridize a similar region of the transcript. The second probe set for this gene showed no significant variation across the 16 chips, reflecting the real biological status of the 16 replicates. We therefore presumed that the first probe set against Filamin 1 could be considered an outlier probe set within the HG-U133A array, whose results might be attributable to random variation.
The rate of false discovery of a significant biological change in gene expression, P <0.0001, would be 1 false positive gene in 10 000 genes. For the 22 283 probe sets analyzed in the HG-U133A chip, only one Absent gene was found to be significantly affected by the whole model of sources of variability.
We also explored other methods of identifying differentially expressed genes that could account for the day-to-day variability. Using -adjusted t-tests and SAM on the three different expression value sets obtained with MAS5, MBEI, and RMA, we created a common list of differentially expressed genes containing 10 probe sets. These expression values were significantly higher on the first day than on the second day. Interestingly, these 10 probe sets were part of the 18 probe sets that are designed to detect the set of hybridization exogenous controls recommended by Affymetrix: BioB, BioC, BioD, and Cre(7). The remaining eight exogenous control probe sets were not significantly affected by the day-to-day effect, suggesting that these 10 probe sets might be more sensitive to subtle variations in target concentration. These transcripts are part of a cocktail that is added to the sample at the moment of hybridization. Because that solution was freshly made each day, variation could have been a result of slight pipetting errors at the moment of hybridization mixture preparation. These results are thus in agreement with the expected run-to-run variability. Even more importantly, the fact that only the hybridization exogenous controls were significantly affected by the day-to-day variability when high-quality total RNA samples were analyzed strongly suggests that a platform such as the Affymetrix high-density oligonucleotide microarrays could be extremely useful for clinical applications in cancer and other diseases stratification.
Finally, from these studies we concluded that total RNA samples could be successfully used in microarray experiments when they met rigorous QC criteria that ensure reliable gene expression data. These QC criteria can also be applicable to RNA samples to be analyzed on different microarray platforms, where efficient cDNA and cRNA synthesis steps are required.
|
Acknowledgments |
|---|
|
Footnotes |
|---|
|
References |
|---|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionReferences |
|---|
The following articles in journals at HighWire Press have cited this article:
|
|
 |
|
  X. Ye and R. Lotan Potential misinterpretation of data on differential gene expression in normal and malignant cells in vitro Brief Funct Genomic Proteomic, July 1, 2008; 7(4): 322 - 326. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 C. I. Dumur, M. Lyons-Weiler, C. Sciulli, C. T. Garrett, I. Schrijver, T. K. Holley, J. Rodriguez-Paris, J. R. Pollack, J. L. Zehnder, M. Price, et al. Interlaboratory Performance of a Microarray-Based Gene Expression Test to Determine Tissue of Origin in Poorly Differentiated and Undifferentiated Cancers J. Mol. Diagn., January 1, 2008; 10(1): 67 - 77. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 S. S. Winter, Z. Jiang, H. M. Khawaja, T. Griffin, M. Devidas, B. L. Asselin, and R. S. Larson Identification of genomic classifiers that distinguish induction failure in T-lineage acute lymphoblastic leukemia: a report from the Children's Oncology Group Blood, September 1, 2007; 110(5): 1429 - 1438. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 G. Sadri-Vakili, B. Bouzou, C. L. Benn, M.-O. Kim, P. Chawla, R. P. Overland, K. E. Glajch, E. Xia, Z. Qiu, S. M. Hersch, et al. Histones associated with downregulated genes are hypo-acetylated in Huntington's disease models Hum. Mol. Genet., June 1, 2007; 16(11): 1293 - 1306. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 K. J. Archer and T. Guennel An application for assessing quality of RNA hybridized to Affymetrix GeneChips Bioinformatics, November 1, 2006; 22(21): 2699 - 2701. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 K. J. Archer, C. I. Dumur, S. E. Joel, and V. Ramakrishnan Assessing quality of hybridized RNA in Affymetrix GeneChip experiments using mixed-effects models Biostat., April 1, 2006; 7(2): 198 - 212. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
T. M. Daly, C. M. Dumaual, C. A. Dotson, M. W. Farmen, S. K. Kadam, and R. D. Hockett Precision Profiling and Components of Variability Analysis for Affymetrix Microarray Assays Run in a Clinical Context J. Mol. Diagn., August 1, 2005; 7(3): 404 - 412. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 S. Imbeaud, E. Graudens, V. Boulanger, X. Barlet, P. Zaborski, E. Eveno, O. Mueller, A. Schroeder, and C. Auffray Towards standardization of RNA quality assessment using user-independent classifiers of microcapillary electrophoresis traces Nucleic Acids Res., March 30, 2005; 33(6): e56 - e56. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
