- Research article
- Open Access
Sources of variability and effect of experimental approach on expression profiling data interpretation
© Bakay et al; licensee BioMed Central Ltd. 2002
- Received: 27 November 2001
- Accepted: 31 January 2002
- Published: 31 January 2002
We provide a systematic study of the sources of variability in expression profiling data using 56 RNAs isolated from human muscle biopsies (34 Affymetrix MuscleChip arrays), and 36 murine cell culture and tissue RNAs (42 Affymetrix U74Av2 arrays).
We studied muscle biopsies from 28 human subjects as well as murine myogenic cell cultures, muscle, and spleens. Human MuscleChip arrays (4,601 probe sets) and murine U74Av2 Affymetrix microarrays were used for expression profiling. RNAs were profiled both singly, and as mixed groups. Variables studied included tissue heterogeneity, cRNA probe production, patient diagnosis, and GeneChip hybridizations. We found that the greatest source of variability was often different regions of the same patient muscle biopsy, reflecting variation in cell type content even in a relatively homogeneous tissue such as muscle. Inter-patient variation was also very high (SNP noise). Experimental variation (RNA, cDNA, cRNA, or GeneChip) was minor. Pre-profile mixing of patient cRNA samples effectively normalized both intra- and inter-patient sources of variation, while retaining a high degree of specificity of the individual profiles (86% of statistically significant differences detected by absolute analysis; and 85% by a 4-pairwise comparison survival method).
Using unsupervised cluster analysis and correlation coefficients of 92 RNA samples on 76 oligonucleotide microarrays, we found that experimental error was not a significant source of unwanted variability in expression profiling experiments. Major sources of variability were from use of small tissue biopsies, particularly in humans where there is substantial inter-patient variability (SNP noise).
- Duchenne Muscular Dystrophy
- Gene Expression Change
- Duchenne Muscular Dystrophy
- Tissue Heterogeneity
- Individual Profile
Expression profiling is an emerging experimental method whereby RNA accumulation in cells and tissues can be assayed for many thousands of genes simultaneously in a single experiment. There are two common experimental platforms for expression profiling; redundant oligonucleotide arrays (Affymetrix GeneChips) , and spotted cDNA microarrays [2–4]. The Affymetrix GeneChips have the inherent advantages of redundancy, specificity, and transportability; there are typically 30–40 oligonucleotide probes (features) designed against each gene tested by the array, with paired perfect-match and mismatch probes, with standardized factory synthesis of arrays [5, 6]. The uniform nature of the arrays permits databasing of individual profiles, which facilitates comparison of data generated by different laboratories.
Expression profiling has led to dramatic advances in understanding of yeast biology, where homogeneous cultures can be grown and exposed to timed environmental variables [7–12]. Such studies have led to the rapid assignment of function to a large number of anonymous gene sequences. Large-scale expression profiling studies of tissues from higher vertebrates are more challenging, due to the higher complexity of the genome, larger related gene families, and incomplete genomic resources. Nevertheless, DNA microarrays have been successfully applied in the analysis of aging and caloric restriction  and pulmonary fibrosis . And many publications, particularly on cancer, have appeared [14–19]. Affymetrix has recently announced the availability of the U133 GeneChip series with 33,000 well-characterized human genes mined from genomic sequence. The nearly complete ascertainment of genes in the human genome should make expression-profiling studies of human tissues particularly powerful. However, identification of the sources of experimental variability, and knowledge of the relative contribution of variation from each source, is critical for appropriate experimental design in expression profiling experiments.
Mills and Gordon recently studied the relative contribution of experimental variability of probe production on the reproducibility of microarray results using mixed murine tissue RNA on Affymetrix Mu11K GeneChips . In their study, the same RNA preparation was used as a template for distinct cDNA/cRNA amplifications and hybridizations. An additional variable studied was the effect of different laboratories processing the same RNAs. The authors found relatively poor concordance between duplicate arrays, with an average of 12% increase/decrease calls between the same RNA processed in parallel and hybridized to two Mu11K-A microarrays. The authors concluded that there was substantial experimental variability in the experimental procedure, necessitating extensive filtering and large numbers of arrays to detect accurate gene expression changes (LUT: look-up tables) . In our laboratory, we have processed over 1,200 Affymetrix arrays, and have found significantly higher experimental reproducibility (R2 = 0.979 for new generation U74A version 2 murine arrays or human U95 series, see Result and Discussion). In addition, a recent publication of a single human patient, where RNA was prepared from two distinct breast tumors, and placed on duplicate U95A GeneChips (four chips total) found a very low degree of experimental variability between microarrays (R2 = 0.995), and between the two tumors (R2 = 0.987) . The marked differences in experimental variability between laboratories could be due to different quality control protocols (see http://microarray.cnmcresearch.org), newer more robust Affymetrix arrays now available (murine Mu11K versus U74A version 2 and new generation human U95 series), use of more recent algorithms for data interpretation, or due to more consistent processing of RNA, cDNA, and cRNA in the same laboratory.
The previous studies did not systematically address the reproducibility of GeneChip hybridization (e.g. the same biotinylated cRNA on two different microarrays). In addition to lingering questions concerning variability due to specific experimental procedures, there are other possible sources of variability that have not yet been investigated, specifically tissue heterogeneity and inter-individual variation. The latter two sources of variability are particularly important in human expression profiling studies. The study of human tissues often involves the use of tissue biopsies, where a relatively limited region of an organ is sampled. Tissue heterogeneity and sampling error might be expected to introduce significant variability in expression profiles. Second, tissues may derive from individuals from different ethnic backgrounds; humans are highly outbred, leading to the potential of significant polymorphic noise (herein called "SNP noise") between individuals unrelated to the disease or variable under study. SNP noise also exists between different inbred mouse strains, and some experiments have normalized this effect by breeding the same mutation on different strains, and profiling each individually . Knowledge of the relative effect of each experimental, tissue, and patient variable on expression profiling results in humans is important, so that appropriate experimental designs can be employed.
We recently reported the design and production of a highly redundant oligonucleotide microarray for analysis of human muscle biopsies (Borup et al. submitted). This MuscleChip contains 4,601 probe sets corresponding to 3,369 distinct genes and ESTs expressed in human muscle. Each probe set contains between 16 to 40 oligonucleotides, such that the number of specific oligonucleotide probes on the array was 138,000.
Here, we utilize this MuscleChip to investigate the relative significance of variables affecting expression profiling data and interpretation. Specifically, we studied the correlation coefficients of profiles considering the following variables: 1. variation due to probe production (same RNA); 2. variation due to the microarray itself (same cRNA on different GeneChips); 3. tissue heterogeneity (different regions of the same muscle biopsy); 4. inter-patient variability (SNP noise); 5. diagnosis (underlying pathological variable); and 6. patient age.
We have recently reported generation of expression profiling results using mixed patient samples . Our hypothesis was that mixing of RNA samples from multiple regions of muscle biopsies, and from multiple patients matched for most variables (disease, age, sex), would effectively normalize both intra-patient variability (tissue heterogeneity), and inter-patient variability (SNP noise; e.g. normal human polymorphic variation unrelated to the primary defect). Here, we test this hypothesis directly, and show that sample mixing does indeed result in relatively high sensitivity and specificity for gene expression changes that would be detected by many individual expression profiles. Thus, sample mixing appears to be an appropriate first-pass method to obtain the most significant expression changes, while using small numbers of arrays.
Patient data and characteristics of 34 MuscleChip expression profiles.
Individual or Mixed
Stage of histopathology
% Present Calls
% Diff Calls Paired samples
Four comparisons > 2-fold changes relative to controls 1a, 1b
6 to 9
5 patients biopsies, cRNAs mixed
5 to 6
5 patients biopsies, cRNAs mixed
10 to 12
5 patients biopsies, cRNAs mixed
6 to 9
5 normal biopsies, cRNAs mixed
5 to 12
3 normal biopsies, cRNAs mixed
4 to 13
3 normal biopsies, cRNAs mixed
Among the 4,601 probe sets on the Affymetrix custom muscle microarray, we found a consistent percentage of "present" calls for each of the 34 cRNA samples tested (Duchenne dystrophy, 28 arrays, 48.2% ± 6.1%; controls 6 arrays, 53.3% ± 1.4%). To test for inter-array variability, two different hybridization solutions were applied to duplicate arrays, and correlation coefficients determined. A high correlation coefficient was found in this analysis, suggesting that inter-array variability of the MuscleChip used was a relatively minor variable (Patient 3 a and 3a-duplicate R2 = 0.96 and percent shared [No Change (NC)] calls by Microarray Suite software was 99%; Patient 3b and 3b-duplicate R2 = 0.98 and percent NC was 98%; Table 1). The high reproducibility of Affymetrix array results is consistent with other data in our laboratory, and from previously published data [6, 21, 23, 24], and shows that experimental variability associated with hybridization and scanning of highly redundant oligonucleotide GeneChips is not a major source of experimental variability.
To analyze the impact of intra-patient variability (tissue heterogeneity), inter-patient variability (polymorphic noise in outbred populations), and the effect of sample mixing on the sensitivity of detection of gene expression differences between patient groups, we conducted a series of individual and mixed profiling (Table 1). Muscle biopsies from five 4–6 yr old DMD patients, and five 10–12 yr old patients were selected, each biopsy split into two parts, and RNA isolated independently from each of the 20 biopsy fragments. For these ten DMD patients, the two different regions of the same biopsy were expression profiled both individually (20 profiles), and also mixed into four pools where each pool originated from distinct RNA samples (Table 1). The resulting profiles were also compared to previously reported mixed 6–9 yr old DMD patient cRNAs, and mixed 6–9 yr old control cRNAs , as mentioned above.
The most common strategy for interpreting Affymetrix microarray data is to use two profile comparisons, with an arbitrary threshold for "significant fold-change" in expression levels. Typically, multiple arrays are compared, with those gene expression changes showing the most consistent fold changes prioritized, although other methods have been reported [13, 22, 26, 27]. To study inter-patient variability, we defined the gene expression changes surviving four pairwise comparisons with mixed control samples, as we have previously described . Briefly, four comparisons were done by Affymetrix software (eg. DMD 1a versus control 1a; DMD 1a versus control 1b; DMD1b versus control 1a; DMD1b versus control 1b). The four data sets were then compared, with only those gene expression changes that showed >2-fold change in all four comparisons (four comparison survival method). The number of surviving diff calls by this method ranged from 250 to 463 (355 ± 80) (Table 1). Interestingly, those patients showing considerable variation between different regions of the same biopsy did not show a corresponding decrease in the number of gene expression changes surviving the iterative comparisons to controls (Table 1). This suggests (but does not prove) the most significant changes might be shared, independent of tissue variability (see below).
When comparing two different regions of the same biopsy [intra-patient variability], we found widely varying results, depending on the patient studied (Fig. 3). For example, some individual patients showed very closely related profiles that approached the similarity of duplicate arrays on the same cRNA (Fig. 2; profiles 6a, 6b; 10a, 10b). On the other hand, some patients showed very distantly related profiles for two regions of the same biopsy (Fig. 3; profiles 1a, 1b; 4a, 4b; 9a, 9b). Importantly, the variation caused by intra-patient tissue variation often overshadowed all other variables. For example, a profile from DMD patient 9 (9a) clustered with the normal controls, rather than with the other DMD patients (Fig. 3). The histopathology of this patient was noted as being unusually variable in severity prior to expression profiling. Also, unsupervised clustering was unable to group patients of similar ages, despite DMD showing a progressive clinical course. We conclude that intra-patient tissue heterogeneity is a major source of experimental variability in expression profiling, and must be considered in experimental design.
The above findings suggested that both intra-patient variability (tissue heterogeneity) and inter-patient variability (polymorphic noise) had major effects on the expression profiles. One method to control for these sources of noise is to analyze large numbers of profiles, both on multiple patients, and on multiple regions of tissue from each patient. This would allow determinations of p values and statistical significance for a single controlled variable under study (e.g. DMD vs controls). An alternative method is to experimentally normalize these variables through mixing of samples from patient groups; such mixing would be expected to average out both intra- and inter-patient variation. The expectation is that the most significant and dramatic gene expression changes would still be identified, while using many less profiles (and thus a substantial reduction in cost of the analyses).
All 34 profiles (both individual and mixed samples) were again analyzed by unsupervised hierarchical clustering (Fig. 4) . As described above, we scrubbed the profiles to eliminate all genes showing expression levels consistently at or below background hybridization intensities by requiring each gene to show a "Present Call" in one or more of the 34 profiles.
As above, duplicate profiles using the same cRNA hybridization solution on different arrays, whether mixed or individual samples, showed very highly correlated results (very low branch on dendrogram) (Fig. 4; mix 5–6 yrs, mix 10–12 yrs; patient 3a/3a-d; patient 3b/3b-d). As above, this indicates that experimental variability from laboratory procedures or different arrays is a relatively minor factor in interpretation of results. Mixed samples from different regions of the same biopsies showed the same, or only slightly more variation (mixed controls c1, c2, and c3, mixed DMD 6–9 yrs). This showed that sample mixing does indeed average out tissue heterogeneity (intra-patient variability), as well as inter-patient variability. We noted that all of the controls (both male and female) clustered in the same branch of the dendrogram, while the four of the six mixed DMD profiles clustered just one level away from the controls, separately from the other DMD profiles. This analysis suggests that there is considerable variability in the progressive tissue pathology induced by dystrophin deficiency, both within a patient, and between patients.
To test the sensitivity and specificity of sample mixing versus individual profiling, we defined differentially expressed genes using a two group t-test (GeneSpring [28, 29]), comparing all 6 mixed control profiles and the 10 individual 5–6 yr old DMD profiles. Genes were retained that met specific p value thresholds between the two sets of profiles. In parallel, we compared the two corresponding mixed 5–6 yr old DMD profiles to the same 6 mixed control profiles.
Use of t-test measurements is expected to contain significant amounts of noise, due to the very large number of comparisons involved in array studies; a value of p = 0.05 means that as many as 5% of gene expression changes are expected to be identified by "chance", and thereby not reflect true differences between samples. We have previously reported a very simple, yet potentially more stringent method for data analysis of small numbers of expression profiles, using duplicate profiles for control and experimental samples, and then identifying those genes that show consistent changes >2-fold in the four possible pair-wise data comparisons (four comparison survival method) . A similar pair-wise comparison method, using a less stringent average fold-change analysis, was recently reported for muscle from aging and calorie-restricted mouse muscle .
Microarray data analyses have been criticized as being "quite elusive about measurement reproducibility" . This is largely the consequence of the large number of uncontrolled or unknown variables, and the prohibitive cost of isolating and investigating each variable. Here, we report the systematic isolation and study of most variables in microarray experiments using Affymetrix oligonucleotide arrays and human tissue biopsies. We found that all sources of experimental variability were quite minor (microarray R2 = 0.98–0.99; probe synthesis + microarray R2 = 0.98–0.99). On the other hand, tissue heterogeneity (intra-patient variation; Average R2 for 10 patients = 0.92 [0.85 to 0.98]), and differences between individual patients (SNP noise; Average R2 = 0.76 [0.42 to 0.93]) were major sources of variability in expression profiling. Thus, tissue heterogeneity and SNP noise have a high potential to obscure sought after condition-specific gene expression changes, particularly in humans, where tissue samples can be limiting (sampling error), and inter-individual variation often is very large. We have shown that mixing of patient samples effectively normalizes much of the intra- and inter-patient noise, while still identifying the majority of the most significant gene expression changes that would have been detected by larger numbers of individual patient profiles. Our results suggest that stringent yet robust data can be generated by mixing a small number of individuals with a defined condition (n = 5), preferably using different regions of tissue for duplicate arrays. Controls should be similarly processed. The resulting four arrays (2 controls, 2 experimental datasets) should then be subjected to the >2-fold survival method, as previously described . This will yield a stringent set of expression changes that are likely to be verified by larger studies with individual arrays, but at low cost as only four arrays are employed. The preliminary data from just four mixed profiles (two experimental and two control) can then be used to generate functional clusters and pathophysiological models. These preliminary models can then direct more hypothesis-driven experiments, or more extensive expression profiling studies.
Human muscle biopsy samples were diagnostic specimens flash-frozen immediately after surgery in isopentane cooled in liquid nitrogen, with storage in small, airtight, humidified tubes at -80°C until RNA isolation. Duchenne muscular dystrophy patient samples were all shown to have complete lack of dystrophin by immunostaining and/or immunoblot analysis, and were shown to have excellent morphology and preservation of tissue. Controls included groups of males and female (age described in text) that showed no histopathological abnormality, normal dystrophy proteins, and normal serum creatine kinase levels. Biopsy sizes ranged from 50 mg to 2 grams, with approximately 20–30 mg used for RNA isolation (~10–15 micrograms of total RNA). As described in the text, all biopsies had two different regions of the same biopsy expression profiled separately.
Details concerning the murine profiles will be published elsewhere. In this report, we used the murine profiles simply to test the sources of variation during sample preparation prior to hybridization to oligonucleotides.
RNA isolation (Trizol, Gibco BRL), RNA purification (RNAeasy, Qiagen), cDNA synthesis and biotinylated cRNA were all done as per standard protocols provided by Affymetrix Inc. Quality control methods are described on our web site (http://microarray.cnmcresearch.org/pga.htm), with cRNA amplifications of between 5- and 13-fold for each of the samples. Ten micrograms of gel-verified fragmented biotinylated cRNA were hybridized to each MuscleChip or U74A v2 array, and scanning done after biotin/avidin/phycoerythrin amplification. Details on the specific patients studied, and details for each GeneChip (scaling factors, number of present calls, percentage difference calls between each duplicate sample, number of difference calls surviving four pair-wise comparisons of duplicate chips) is provided (Table 1). All profiling data presented here is available on our web site (http://microarray.CNMCResearch.org; data link), as image (.dat), absolute analysis (.chp), and ASCII text conversions of .chp (.txt) for each individual profile (see http://microarray.cnmcresearch.org/pga.htm for file descriptions and use).
Absolute analysis (average difference determinations for each probe set) was done using Affymetrix default parameters. As described in the text, data was analyzed using a variety of methods, including unsupervised nearest-neighbor hierarchical clustering analyses (GeneSpring [28, 29] [Silicon Genetics], and Cluster  [Stanford University]), t-test (GeneSpring) and four-comparison survival method . The Cluster and Tree View software were download from http://rana.lbl.gov and installed on an NT workstation.
We would like to thank Brian S. Tseng and Frank W. Booth (Department of Physiology University of Missouri, Columbia, MO) and Simona lezzi and Vittorio Sartorelli (Muscle Biophysics Section, Laboratory of Physical Biology NIAMS/NIH Bethesda, MD) for allowing to use their data in our analysis.
Drs. Bakay and Chen were supported by post-doctoral fellowships from the Stichting-Porticus Foundation. Supported in part by grants from the National Institutes of Health (5RO1 NS29525-10; and a "Programs in Genomic Applications" from NHLBI [UO1 HL66614-01]) to EPH.
- Lipshutz RJ, Fodor SP, Gingeras TR, Lockhart DJ: High density synthetic oligonucleotide arrays. Nature Genetics 1999, 21: 20–24. 10.1038/4447View ArticlePubMedGoogle Scholar
- Duggan DJ, Bittner M, Chen Y, Meltzer P, Trent JM: Expression profiling using cDNA microarrays. Nature Genetics 1999, 21: 10–14. 10.1038/4434View ArticlePubMedGoogle Scholar
- Brown PO, Botstein D: Exploring the new world of the genome with DNA microarrays. Nature Genetics 1999, 21: 33–37. 10.1038/4462View ArticlePubMedGoogle Scholar
- Cheung VG, Morley M, Aguilar F, Massimi A, Kucherlapati R, Childs G: Making and reading microarrays. Nature Genetics 1999, 21: 15–19. 10.1038/4439View ArticlePubMedGoogle Scholar
- McGall G, Labadie J, Brock P, Wallraff G, Nguyen T, Hinsberg W: Light-directed synthesis of high-density oligonucleotide arrays using semicomductor photoresists. Proc. Natl. Acad. Sci. USA 1996, 93: 13555–13560. 10.1073/pnas.93.24.13555PubMed CentralView ArticlePubMedGoogle Scholar
- Lockhart DJ, Dong H, Byrne MC, Follettie MT, Gallo MV, Chee MS, Mittmann M, Wang C, Kobayashi M, Horton H, Brown EL: Expression monitoring by hybridization to high-density oligonucleotide arrays. Nat. Biotechnol. 1996, 14: 1675–1680.View ArticlePubMedGoogle Scholar
- Winzeler EA, Richards DR, Conway AR, Goldstein AL, Kalman S, McCullough MJ, McCusker JH, Stevens DA, Wodicka L, Lockhart DJ, Davis RW: Direct allelic variation scanning of the yeast genome. Science 1998, 281: 1194–1197. 10.1006/bbrc.2001.4484View ArticlePubMedGoogle Scholar
- Spellman PT, Sherlock G, Zhang MQ, Iyer VR, Anders K, Eisen MB, Brown PO, Botstein D, Futcher B: Comprehensive identification of cell cycle-regulated genes of the yeast Saccharomyces cerevisiae by microarray hybridization. Mol Biol Cell 1998, 9: 3273–3297.PubMed CentralView ArticlePubMedGoogle Scholar
- Gasch AP, Spellman PT, Kao CM, Carmel-Harel O, Eisen MB, Storz G, Botstein D, Brown PO: Genomic expression programs in the response of yeast cells to environmental changes. Mol Biol Cell 2000, 11: 4241–4257.PubMed CentralView ArticlePubMedGoogle Scholar
- Sudarsanam P, Iyer VR, Brown PO, Winston F: Whole-genome expression analysis of snf/swi mutants of Saccharomyces cerevisiae. Proc. Natl. Acad. Sci. USA 2000, 97: 12369–12374. 10.1073/pnas.050407197View ArticleGoogle Scholar
- Cavalieri D, Townsend JP, Hartl DL: Manifold anomalies in gene expression in a vineyard isolate of Saccharomyces cerevisiae revealed by DNA microarray analysis. Proc. Natl. Acad. Sci. USA 2000, 97: 12369–12374. 10.1073/pnas.210395297PubMed CentralView ArticlePubMedGoogle Scholar
- Ogawa N, DeRisi J, Brown PO: New components of a system for phosphate accumulation and polyphosphate metabolism in Saccharomyces cerevisiae revealed by genomic expression analysis. Mol Biol Cell 2000, 11: 4309–4321.PubMed CentralView ArticlePubMedGoogle Scholar
- Lee CK, Klopp RG, Weindruch R, Prolla TA: Gene expression profile of aging and its retardation by caloric restriction. Science 1999, 285: 1390–1393. 10.1126/science.285.5432.1390View ArticlePubMedGoogle Scholar
- Kaminski N, Allard JD, Pittet JF, Zuo F, Griffiths MJ, Morris D, Huang X, Sheppard D, Heller RA: Global analysis of gene expression in pulmonary fibrosis reveals distinct programs regulating lung inflammation and fibrosis. Proc. Natl. Acad. Sci. USA 2000, 97: 1778–1783. 10.1073/pnas.97.4.1778PubMed CentralView ArticlePubMedGoogle Scholar
- Golub TR, Slonim DK, Tamayo P, Huard C, Gaasenbeek M, Mesirov JP, Coller H, Loh ML, Downing JR, Caligiuri MA, Bloomfield CD, Lander ES: Molecular classification of cancer: class discovery and class prediction by gene expression monitoring. Science 1999, 286: 531–537. 10.1016/S0378-4371(00)00404-0View ArticlePubMedGoogle Scholar
- Perou CM, Jeffrey SS, van de Rijn M, Rees CA, Eisen MB, Ross DT, Pergamenschikov A, Williams CF, Zhu SX, Lee JC, Lashkari D, Shalon D, Brown PO, Botstein D: Distinctive gene expression patterns in human mammary epithelial cells and breast cancers. Proc. Natl. Acad. Sci. USA 1999, 96: 9212–9217. 10.1073/pnas.96.16.9212PubMed CentralView ArticlePubMedGoogle Scholar
- Sorlie T, Perou CM, Tibshirani R, Aas T, Geisler S, Johnsen H, Hastie T, Eisen MB, van de Rijn M, Jeffrey SS, Thorsen T, Quist H, Matese JC, Brown PO, Botstein D, Eystein Lonning P, Borresen-Dale AL: Gene expression patterns of breast carcinomas distinguish tumor subclasses with clinical implications. Proc. Natl. Acad. Sci. USA 2001, 98: 10869–74. 10.1073/pnas.191367098PubMed CentralView ArticlePubMedGoogle Scholar
- Bittner M, Meltzer P, Chen Y, Jiang Y, Seftor E, Hendrix M, Radmacher M, Simon R, Yakhini Z, Ben-Dor A, Sampas N, Dougherty E, Wang E, Marincola F, Gooden C, Lueders J, Glatfelter A, Pollock P, Carpten J, Gillanders E, Leja D, Dietrich K, Beaudry C, Berens M, Alberts D, Sondak V: Molecular classification of cutaneous malignant melanoma by gene expression profiling. Nature 2000, 406: 536–540. 10.1038/35020115View ArticlePubMedGoogle Scholar
- Alon U, Barkai N, Notterman DA, Gish K, Ybarra S, Mack D, Levine AJ: Broad patterns of gene expression revealed by clustering analysis of tumor and normal colon tissues probed by oligonucleotide arrays. Proc. Natl. Acad. Sci. USA 1999, 96: 6745–6750. 10.1073/pnas.96.12.6745PubMed CentralView ArticlePubMedGoogle Scholar
- Mills JC, Gordon JI: A New approach for filtering noise from high-density oligonucleotide microarray datasets. Nucleic Acids Res 2001, 29: E72–2. 10.1093/nar/29.15.e72PubMed CentralView ArticlePubMedGoogle Scholar
- Unger MA, Rishi M, Clemmer VB, Hartman JL, Keiper EA, Greshock JD, Chodosh LA, Liebman MN, Weber BL: Characterization of adjacent breast tumors using oligonucleotide microarrays. Breast Cancer Res 2001, 3: 336–341. 10.1186/bcr317PubMed CentralView ArticlePubMedGoogle Scholar
- Nadler ST, Stoehr JP, Schueler KL, Tanimoto G, Yandell BS, Attie AD: The expression of adipogenic genes is decreased in obesity and diabetes mellitus. Proc. Natl. Acad. Sci. USA 2000, 97: 11371–11376. 10.1073/pnas.97.21.11371PubMed CentralView ArticlePubMedGoogle Scholar
- Chen Y, Zhao P, Borup R, Hoffman EP: Expression profiling in the muscular dystrophies: identification of novel aspects of molecular pathophysiology. J. Cell Biol. 2000, 151: 1321–1336. 10.1083/jcb.151.6.1321PubMed CentralView ArticlePubMedGoogle Scholar
- Kamb A, Ramaswami M: A simple method for statistical analysis of intensity differences in microarray-derived gene expression data. BMC Biotechnol 2001, 1: 8. 10.1186/1472-6750-1-8PubMed CentralView ArticlePubMedGoogle Scholar
- Eisen MB, Spellman PT, Brown PO, Botstein D: Cluster analysis and display of genome-wide expression patterns. Proc. Natl. Acad. Sci. USA 1998, 95: 14863–14868. 10.1073/pnas.95.25.14863PubMed CentralView ArticlePubMedGoogle Scholar
- Der SD, Zhou A, Williams BR, Silverman RH: Identification of genes differentially regulated by interferon alpha, beta, or gamma using oligonucleotide arrays. Proc. Natl. Acad. Sci. USA 1998, 95: 15623–15628. 10.1073/pnas.95.26.15623PubMed CentralView ArticlePubMedGoogle Scholar
- Hooper LV, Wong MH, Thelin A, Hansson L, Falk PG, Gordon JI: Molecular analysis of commensal host-microbial relationships in the intestine. Science 2001, 291: 881–884. 10.1126/science.291.5505.881View ArticlePubMedGoogle Scholar
- GeneSpring User Manual, version 4.0, March Silicon Genetics 2001.Google Scholar
- GeneSpring Advanced Analysis Techniques, version 4.0, March Silicon Genetics 2001.Google Scholar
- Claverie JM: Computational methods for the identification of differential and coordinated gene expression. Hum Mol Genet 1999, 8: 1821–1832. 10.1093/hmg/8.10.1821View ArticlePubMedGoogle Scholar
This article is published under license to BioMed Central Ltd. This is an Open Access article: verbatim copying and redistribution of this article are permitted in all media for any purpose, provided this notice is preserved along with the article's original URL.