Physiol. Genomics Fuel your research with LabChart
HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS
QUICK SEARCH:   [advanced]


     

Physiol. Genomics 26: 158-162, 2006. First published May 9, 2006; doi:10.1152/physiolgenomics.00313.2005 Free Article
1094-8341/06 $8.00
This Article
Free upon publication Free Article
Right arrow Abstract Freely available
Right arrow Full Text (PDF)
Right arrow Supplementary Material
Right arrow All Versions of this Article:
26/2/158    most recent
00313.2005v1
Right arrow Alert me when this article is cited
Right arrow Alert me if a correction is posted
Right arrow Citation Map
Services
Right arrow Email this article to a friend
Right arrow Similar articles in this journal
Right arrow Similar articles in PubMed
Right arrow Alert me to new issues of the journal
Right arrow Download to citation manager
Citing Articles
Right arrow Citing Articles via ISI Web of Science (3)
Right arrow Citing Articles via Google Scholar
Google Scholar
Right arrow Articles by Liang, S.
Right arrow Articles by Liu, W.
Right arrow Search for Related Content
PubMed
Right arrow PubMed Citation
Right arrow Articles by Liang, S.
Right arrow Articles by Liu, W.
Received 19 December 2005; accepted in final form 4 May 2006.
Physiological Genomics 26:158-162 (2006)
1094-8341/06 $8.00 © 2006 American Physiological Society

Detecting and profiling tissue-selective genes

Shuang Liang, Yizheng Li, Xiaobing Be, Steve Howes and Wei Liu

Bioinformatics, Wyeth Research, Cambridge, Massachusetts


   ABSTRACT
TOP
 ABSTRACT
INTRODUCTION
 METHODS
 RESULTS
 DISCUSSION
 REFERENCES
 
The widespread use of DNA microarray technologies has generated large amounts of data from various tissue and/or cell types. These data set the stage to answer the question of tissue specificity of human transcriptome in a comprehensive manner. Our focus is to uncover the tissue-gene relationship by identifying genes that are preferentially expressed in a small number of tissue types. The tissue selectivity would shed light on the potential physiological functions of these genes and provides an indispensable reference to compare against disease pathophysiology and to identify or validate tissue-specific drug targets. Here we describe a systematic computational and statistical approach to profile gene expression data to identify tissue-selective genes with the use of a more extensive data set and a well-established multiple-comparison procedure with error rate control. Expression data of 35,152 probe sets in 97 normal human tissue types were analyzed, and 3,919 genes were identified to be selective to one or a few tissue types. We presented results of these tissue-selective genes and compared them to those identified by other studies.

tissue selectivity; differential expression; transcription profiling; Tukey; honest significant difference


   INTRODUCTION
TOP
 ABSTRACT
 INTRODUCTION
METHODS
 RESULTS
 DISCUSSION
 REFERENCES
 
THE MAJORITY OF STUDIES using DNA microarray technology have been focused on the detection of differences in gene expression between two selected pathological categories (e.g., tumor vs. nontumor tissue types) or between treated and untreated conditions (1, 5, 21). Recently, some attention has turned to tissue-specific transcription profiling to predict clinical outcome (1, 13, 27), to screen for potential drug targets (29), or to survey gene expression in a panel of normal human tissue types (1, 7, 11, 17, 19, 22, 24). Preferential expression measure (8) and a method based on modified Poisson statistics (17) were developed for expressed sequence tag (EST) and serial analysis of gene expression (SAGE) data. These surveys covered as few as some 4,000 ESTs to as many as over a million tags for SAGE. Other approaches were used in microarray-based expression profiling: for instance, two-tailed Student's t-test and principal component analysis (PCA) were used to analyze 19–40 human tissue types (7, 19); reassociation kinetics was used to assess the global trend of expression in 79 human and 61 mouse tissue types (26); Akaike's information criterion was used to identify expression outliers in 48 mouse tissue types (11), and more recently unsupervised hierarchical cluster analysis was applied to analyze expression using 35 human tissue types (24). Although these studies provide useful expression information, they are limited in the spectrum of normal tissues and analytical approaches to tissue specificity of human transcriptome.

In its strictest sense, tissue specificity is defined as gene expression being exclusive to only one particular tissue type. Unlike tissue specificity, tissue selectivity considers genes whose expression is enriched to one or a few biologically similar tissue types (such as different portions of the digestive track or various brain sections). Tissue selectivity is the focus of this study.

The goal of the present study is to take advantage of a very large set of gene expression data available to us and to discover genes preferentially expressed in selected normal human tissue/cell types through statistical analysis. More than 3,800 (3,995 for U133A chip and 3,827 for U133B chip sets) samples from some 120 normal human tissue types were used to generate expression data using Affymetrix's U133 chip set by GeneLogic. By profiling 35,152 probe sets, which represent >27,000 human genes, we identified 3,919 genes preferentially expressed in a small number of tissue types. This list includes novel genes as well as those previously identified.


   METHODS
TOP
 ABSTRACT
 INTRODUCTION
 METHODS
RESULTS
 DISCUSSION
 REFERENCES
 
Selection of probe sets.
The DNA microarray platform (U133A and U133B) used in this study is based on Affymetrix technology. Only singleton (probe sets with suffix "_at") or similar (probe sets with suffix "_s_at") probe sets were chosen for analysis (see http://www.affymetrix.com for probe set definition). Probe sets for microarray experiment control (total 136) and those far from the 3'-end of transcripts (total 2,800) were excluded. A total of 38,400 probe sets from Affymetrix's U133A (18,454) and U133B (19,956) chip sets were preselected this way. Duplicated probe sets or probe sets containing only "absent" calls were also excluded before subsequent analysis. This resulted in a total of 35,152 probe sets (17,275 for U133A and 17,877 for U133B), which represent over ~22,000 annotated genes and many other hypothetical genes (~5,000) from the human genome.

Selection of tissue types and samples.
The expression data used in the present study were derived from BioExpress database, which was licensed from GeneLogic (http://www.genelogic.com). More than 3,800 human samples covering over 120 tissue types were used to survey the selected probe sets. The pathology of these samples was classified as "normal" by GeneLogic. Of the 122 tissue types, 26 are from different Brodmann's areas, mainly the histological components of the cerebral cortex. Our preliminary studies (data not shown) found no evidence of significant differential expression among the Brodmann's areas (see DISCUSSION). This observation led us to regroup the 26 tissue types associated with Brodmann's areas as a single tissue type. Filters were applied to remove samples with ß-actin and GAPDH 5'-to-3' ratio <0.4 or tissue groups whose sample sizes were <3. Finally, there remained 97 tissue types chosen (total of 3,995 samples for U133A and total of 3,827 samples for U133B).

Expression data were retrieved and grouped by tissue type for each probe set. Then log2 transformation was applied to stabilize data variance (20). Erroneous data (i.e., data with not available/not a number/infinity values) were not included in subsequent analyses.

Outlier detection.
We applied an outlier-filtering scheme to remove outlying sample reads on a tissue-by-tissue basis for each probe set after data were log2 transformed. This was done similarly to the influence function procedure proposed by Singh and Nocerino (25), to find groupwise robust estimates (mean and standard variation) of expression of a given probe set in samples of each tissue type. We considered expression reads of a given probe set beyond three times of the robust estimate of standard deviation from the robust estimate of mean in each tissue type as outliers.

Statistical analysis.
Analytically, searching for tissue-selective genes amounts to comparing gene expression over many tissue types. This constitutes a classical multiple-comparison problem. Tukey-Kramer's honest significant difference (HSD) test was used to determine the tissue types in which a gene expressed selectively. It incorporates an error correction for pairwise multiple comparisons to deal with mildly unbalanced designs (18, 30, 31). In practice, there are N = k*(k – 1)/2 pairwise comparisons where k is the number of tissue types used. Implementation of Tukey-Kramer's HSD function generated an adjusted probability for ith comparison (adjusted) (pi) value as well as a Qi value (i = 1 to N) for each pair of comparison, which could subsequently be used for statistical inference (see below). Each Qi value was computed as absolute difference of two group means over the sample size-adjusted standard deviation. To control the error rate resulting from multiple testing (i.e., tens of thousands of HSD tests), there was a need to produce a single probability for each HSD test (pQ) value for each HSD test. This was done by first defining an enrichment score (ES) using the HSD-generated Q values:

Formula
where N is the total number of pairwise comparisons [N = k(k – 1)/2 for k tissue types]. ES was designed to take into account of Q values from all pairwise comparisons in one HSD test to represent tissue selectivity of a gene. The higher the value of ES for a gene, the more selective it would be. Permutations of tissue labels were done 1,000 times to derive an estimated pQ value from each ES. The false discovery rate (FDR) was set to 0.001 to cap the error rate from the multiple tests, using pQ values adjusted based on a method from Ref. 3.

To identify tissues in which a gene is selective, mean tissue expression was ranked from largest to smallest for each probe set. The adjusted pi values ({alpha} = 0.05) generated from HSD were used to count the number of significant pairwise comparisons for each of the top 10 tissue types. To set a proper level of stringency in tissue selectivity, we defined several filters below. First, for a tissue to be considered as selective (preferred) for a gene, the absolute majority (91 out of 97) of all pairwise comparisons involving such tissue (i.e., a given tissue against all other tissues) must have their adjusted pi values below the set {alpha}-level. Second, the number of selected tissue types in which a gene expressed differentially must be a small number for a gene to be considered as tissue selective. This was set to six tissue types (out of 97) in our case.


   RESULTS
TOP
 ABSTRACT
 INTRODUCTION
 METHODS
 RESULTS
DISCUSSION
 REFERENCES
 
Identifying differential gene expression.
There is a need to survey as many human genes as possible to classify genes based on expression profiles across a panel of different normal human tissue types. We carried out a systematic search for genes expressed preferentially in certain tissue types. More than 3,800 samples from 97 normal tissue types were used to profile the expression of about 35,000 probe sets (see METHODS). A comparison summarizing several studies is shown in Table 1 (see description in next section).


View this table:
[in this window]
[in a new window]
  		Table 1. Comparison of several studies

 
Tissue-selective genes were found in 95 of 97 tissue surveyed. With multiplicity (i.e., one gene expressed in multiple tissue types or multiple probe sets assigned to one gene) included, Table 2 lists the gene counts (total ~8,000) found in each of these tissue types. When multiplicity was discounted, the actual number of unique genes identified to be tissue selective is estimated to be 3,919 (see Supplementary data file 2; the online version of this article contains supplemental data), or ~10% of total genes predicted by the human genome (16, 28). This represents on average about two tissue types per gene as selectivity. Many genes (~80%) we identified are known to play roles in the selected tissue types based on their annotated gene ontology terms (2), whereas ~20% other genes are much less characterized or have not been identified as tissue-selective genes previously.


View this table:
[in this window]
[in a new window]
  		Table 2. Gene-tissue selectivity counts

 
When reorganizing the resulting counts of gene-tissue pairs based on organ systems, we found that the circulatory system, neural system, digestive system, and immune system combined account for ~70% of the counts, followed by reproductive system (9.3%) and excretory system (8%) (Fig. 1). The various numbers of selective genes reflect the needs to support different functional and physiological complexity of each organ system, thus suggesting a link between functional gene expression and physiology of tissue, cell types, or organs.


Figure 1
View larger version (18K):
[in this window]
[in a new window]
  		Fig. 1. Overview of tissue-selective gene. Gene-tissue counts are organized by organ systems to illustrate the proportion of selective genes per organ system.

 
Information on biological process, pathway, subcellular localization, and molecular function for each gene identified, as part of the annotation efforts by the Gene Ontology Consortium (2), has facilitated our exploration. The annotation helped us to evaluate the correlation between tissue selectivity and expected functionality of genes identified in our study, especially those with known important biological functions. For example, among liver-selective genes, many are important for metabolism (see Table 1 in supplementary data file 1). This is expected based on the known biological and physiological functions of the liver. Thus the preferential gene expression in normal tissue, cell types, or organs correlates well with their expected physiological roles. The gene annotation for many of the 3,919 genes reveals that they are involved in at least 1,810 different biological processes and 690 different pathways (supplementary data file 2). The numbers might be bigger since many identified genes on our list are poorly characterized or simply have no known functions described.

Comparative study.
In Table 1, we compared results from several studies of gene expression profiling based on DNA microarray technology. In addition to difference in data coverage and number of genes identified, each study used a different approach. The Student t-test used by Hsiao et al. (7) did not consider the error rate associated with multiple comparisons. This could potentially augment the number of false positives. The PCA test by Misra et al. (19) has limited discriminating power when it comes to dealing with large number of tissue types. In that study, only genes specific to the liver, muscle, and brain were separated. The fold-change approach by (22), though commonly used, is not statistically rigorous. We have adapted Tukey's HSD test, which is rigorous and appropriate for multiple comparisons.

Five previous studies on tissue-selective genes have identified a subsets of genes highly expressed in several tissue groups including muscle, heart, liver, and prostate (7, 8, 17, 19, 22). The Venn diagrams (see B in Supplementary data file 1) compare the result of our study to those of the five studies. As can be seen in the Venn diagrams, the between-study agreement is quite significant for the liver and prostate groups. The largest discrepancy comes from the muscle group. The second group with a large discrepancy is the heart group. It is interesting to note that EST-based studies (8, 17) usually generate a higher discrepancy when compared with our results. This discrepancy might be the result of higher occurrence of false positives as sensitivity of EST-based methods depends heavily on sampling and representation of the data used (4).


   DISCUSSION
TOP
 ABSTRACT
 INTRODUCTION
 METHODS
 RESULTS
 DISCUSSION
REFERENCES
 
Data from DNA microarray experiments have made it possible to survey a large number of genes based on differential expression patterns across a large panel of normal human tissue and cell types. We have shown the utility of a tissue-centric approach in gene expression profiling. Compared with previous works, we used a large data set from a commercial source and a well-established multiple comparison procedure. This approach has generated a list of 3,919 genes of high tissue selectivity. Many genes (roughly 80%) identified have been characterized and have a good correlation to tissue function. Novel genes or genes with less characterization represent ~20% of all detected genes.

In identifying biologically important genes, we focused on genes whose expression is enriched or selective to a small number of tissue types. The rationale is that many genes expressed preferentially in related tissue types, and the inclusion of these genes, in addition to those with exclusive expression preference (i.e., specific), would enlarge our scope of investigation and produce a list of candidates with significant potentials. For instance, a number of genes identified to be muscle selective were expressed selectively in both skeletal muscles and heart muscles. The presence in two different muscle types suggests they might be important factors for general muscle functions. In an analysis of genes selective to fat tissue types (adipose and omentum) from the same data source, a separate study has identified a family of genes (among them is adiponutrin) involved in fat metabolism common in different species (12). This good correlation supports and validates our approach.

Data with large overall variation can make test results hard to interpret and even render the results useless. Sample size, data distribution, and technical factors can also affect the outcome of our result. We addressed some of the problems we have control of through data preprocessing, log2 transformation, and outlier exclusion. An outlier detection scheme applied at the tissue type level to exclude outlying samples resulted in ~20% more tissue-selective genes identified (from ~3,200 to ~3,900 before and after application, respectively). This filtering is important to minimize the consequences of the potential violation of assumption of homogeneity of variance, equal group size, and normal distribution in the test used in the subsequent step. To evaluate the impact of the outlier-filtering scheme, we recorded the frequency (F) of a sample being labeled as outlier by the probe sets used (T) on each chip set. Then we define the odds of a sample being called an outlier as: F/T. Our analysis showed that 95% of samples labeled as outliers have odds of <0.0535 for those run on U133A and 0.0492 for those run on U133B (see D in Supplementary data file 1 for histograms). This suggests that the outlier-removal scheme would not alter the number of samples in the analysis in any significant way. Considering the benefit of using it, we opted to use the scheme as a useful filter.

The expression data from GeneLogic are not probe-level intensity reads; instead, they are preprocessed by MAS5. Therefore, many of the recently developed methods [i.e., robust multichip analysis (RMA), model-based expression index (MBEI), and RMA based on GC content] could not be applied (9, 10, 14, 15). Given this fact, we believe the massive data we have might overcome part, if not all, of the problems associated with MAS5 normalization. Recent studies have also voiced the concerns about the design of Affymetrix's probe set. These problems seem to, in part, result from evolving gene/transcript definition and annotation (6). Because we have no access to the probe-level data, we are limited to what can be done to deal with these problems. Instead, we put an effort on higher-level annotation to justify our result.

In our initial study, we started with 121 tissues types available through GeneLogic to cover as many tissue types as possible. However, we noticed that a large group of Brodmann's areas did not produce any selective genes. This was attributed to the tissue selectivity cutoff (about six tissues) and the number of Brodmann's areas (total 26). Further study showed that genes selective to any Brodmann's areas tend to upregulate in the whole group (i.e., no significant difference could be detected among Brodmann's areas). Subsequently, we regrouped all Brodmann's areas as a single tissue type. This resulted in ~150 unique genes selective to Brodmann's areas as a group. Our initial attempt to use confidence intervals generated from HSD failed to address the problem associated with multiple testing. Hence, we defined an enrichment score within the framework of HSD to produce a single pQ value for each HSD test, which could be used in FDR correction. We also took advantage of the pairwise adjusted pi values generated by HSD to identify tissue types for selective genes.

Genes expressed differentially in selected tissue or cell types often share similar expression patterns. It is tempting to hypothesize that subsets of tissue-selective or tissue-specific genes were subjected to common regulation. By investigating the gene-gene relationship (e.g., looking for common upstream 5'-regulatory elements of a group of coexpressed genes), we might be able to gain more insights into the regulation and function of these genes (23). Furthermore, we might use tissue selectivity to direct and test our hypotheses on novel genes or those with very limited information available to better our understanding of the roles they play.


   ACKNOWLEDGMENTS
 
We thank Andrew Hill for providing data access and technical discussion and Lyndon Hicks for technical support on our computer cluster. We also thank the anonymous reviewer for pointing out the multiple testing corrections. This work was funded by Wyeth's postdoctoral program (to S. Liang).


   FOOTNOTES
 
Article published online before print. See web site for date of publication (http://physiolgenomics.physiology.org).

Address for reprint requests and other correspondence: W. Liu, Bioinformatics, Wyeth Research, 35 Cambridge Park Dr., Cambridge, MA 02140 (e-mail: wliu{at}wyeth.com).


   REFERENCES
TOP
 ABSTRACT
 INTRODUCTION
 METHODS
 RESULTS
 DISCUSSION
 REFERENCES
 

  1. Amatschek S, Koenig U, Auer H, Steinlein P, Pacher M, Gruenfelder A, Dekan G, Vogl S, Kubista E, Heider KH, Stratowa C, Schreiber M, and Sommergruber W. Tissue-wide expression profiling using cDNA subtraction and microarrays to identify tumor-specific genes. Cancer Res 64: 844–856, 2004.[Abstract/Free Full Text]
  2. Ashburner M, Ball CA, Blake JA, Botstein D, Butler H, Cherry JM, Davis AP, Dolinski K, Dwight SS, Eppig JT, Harris MA, Hill DP, Issel-Tarver L, Kasarskis A, Lewis S, Matese JC, Richardson JE, Ringwald M, Rubin GM, and Sherlock G. Gene ontology: tool for the unification of biology. The Gene Ontology Consortium. Nat Genet 25: 25–29, 2000.[CrossRef][ISI][Medline]
  3. Benjamini Y and Hochberg JR. Controlling the false discovery rate: a practical and powerful approach to multiple testing. J Royal Statistical Society Series B 57: 289–300, 1995.
  4. Carulli JP, Artinger M, Swain PM, Root CD, Chee L, Tulig C, Guerin J, Osborne M, Stein G, Lian J, and Lomedico PT. High throughput analysis of differential gene expression. J Cell Biochem Suppl 30–31: 286–296, 1998.
  5. Culhane AC, Perriere G, Considine EC, Cotter TG, and Higgins DG. Between-group analysis of microarray data. Bioinformatics 18: 1600–1608, 2002.[Abstract/Free Full Text]
  6. Dai M, Wang P, Boyd AD, Kostov G, Athey B, Jones EG, Bunney WE, Myers RM, Speed TP, Akil H, Watson SJ, and Meng F. Evolving gene/transcript definitions significantly alter the interpretation of GeneChip data. Nucleic Acids Res 33: e175, 2005.
  7. Hsiao LL, Dangond F, Yoshida T, Hong R, Jensen RV, Misra J, Dillon W, Lee KF, Clark KE, Haverty P, Weng Z, Mutter GL, Frosch MP, Macdonald ME, Milford EL, Crum CP, Bueno R, Pratt RE, Mahadevappa M, Warrington JA, Stephanopoulos G, and Gullans SR. A compendium of gene expression in normal human tissues. Physiol Genomics 7: 97–104, 2001.[Abstract/Free Full Text]
  8. Huminiecki L, Lloyd AT, and Wolfe KH. Congruence of tissue expression profiles from Gene Expression Atlas, SAGEmap and TissueInfo databases. BMC Genomics 4: 31, 2003.[CrossRef][Medline]
  9. Irizarry RA, Bolstad BM, Collin F, Cope LM, Hobbs B, and Speed TP. Summaries of Affymetrix GeneChip probe level data. Nucleic Acids Res 31: e15, 2003.
  10. Irizarry RA, Hobbs B, Collin F, Beazer-Barclay YD, Antonellis KJ, Scherf U, and Speed TP. Exploration, normalization, and summaries of high density oligonucleotide array probe level data. Biostatistics 4: 249–264, 2003.[Abstract]
  11. Kadota K, Nishimura S, Bono H, Nakamura S, Hayashizaki Y, Okazaki Y, and Takahashi K. Detection of genes with tissue-specific expression patterns using Akaike's information criterion procedure. Physiol Genomics 12: 251–259, 2003.[Abstract/Free Full Text]
  12. Lake AC, Sun Y, Li JL, Kim JE, Johnson JW, Li D, Revett T, Shih HH, Liu W, Paulsen JE, and Gimeno RE. Expression, regulation, and triglyceride hydrolase activity of Adiponutrin family members. J Lipid Res 46: 2477–2487, 2005.[Abstract/Free Full Text]
  13. Lapointe J, Li C, Higgins JP, van de Rijn M, Bair E, Montgomery K, Ferrari M, Egevad L, Rayford W, Bergerheim U, Ekman P, DeMarzo AM, Tibshirani R, Botstein D, Brown PO, Brooks JD, and Pollack JR. Gene expression profiling identifies clinically relevant subtypes of prostate cancer. Proc Natl Acad Sci USA 101: 811–816, 2004.[Abstract/Free Full Text]
  14. Li C and Hung Wong W. Model-based analysis of oligonucleotide arrays: model validation, design issues and standard error application. Genome Biol 2: RESEARCH0032, 2001.[Medline]
  15. Li C and Wong WH. Model-based analysis of oligonucleotide arrays: expression index computation and outlier detection. Proc Natl Acad Sci USA 98: 31–36, 2001.[Abstract/Free Full Text]
  16. McPherson JD, Marra M, Hillier L, Waterston RH, Chinwalla A, Wallis J, Sekhon M, Wylie K, Mardis ER, Wilson RK, Fulton R, Kucaba TA, Wagner-McPherson C, Barbazuk WB, Gregory SG, Humphray SJ, French L, Evans RS, Bethel G, Whittaker A, Holden JL, McCann OT, Dunham A, Soderlund C, Scott CE, Bentley DR, Schuler G, Chen HC, Jang W, Green ED, Idol JR, Maduro VV, Montgomery KT, Lee E, Miller A, Emerling S, Kucherlapati Gibbs R, Scherer S, Gorrell JH, Sodergren E, Clerc-Blankenburg K, Tabor P, Naylor S, Garcia D, de Jong PJ, Catanese JJ, Nowak N, Osoegawa K, Qin S, Rowen L, Madan A, Dors M, Hood L, Trask B, Friedman C, Massa H, Cheung VG, Kirsch IR, Reid T, Yonescu R, Weissenbach J, Bruls T, Heilig R, Branscomb E, Olsen A, Doggett N, Cheng JF, Hawkins T, Myers RM, Shang J, Ramirez L, Schmutz J, Velasquez O, Dixon K, Stone NE, Cox DR, Haussler D, Kent WJ, Furey T, Rogic S, Kennedy S, Jones S, Rosenthal A, Wen G, Schilhabel M, Gloeckner G, Nyakatura G, Siebert R, Schlegelberger B, Korenberg J, Chen XN, Fujiyama A, Hattori M, Toyoda A, Yada T, Park HS, Sakaki Y, Shimizu N, Asakawa S, Kawasaki K, Sasaki T, Shintani A, Shimizu A, Shibuya K, Kudoh J, Minoshima S, Ramser J, Seranski P, Hoff C, Poustka A, Reinhardt R, Lehrach H; International Human Genome Mapping Consortium. A physical map of the human genome. Nature 409: 934–941, 2001.[CrossRef][Medline]
  17. Megy K, Audic S, and Claverie JM. Heart specific genes revealed by EST sampling. Genome Biol 3: PREPRINT0008, 2002.[Medline]
  18. Miller RG. Simultaneous Statistical Inference. New York: Springer-Verlag, 1981.
  19. Misra J, Schmitt W, Hwang D, Hsiao LL, Gullans S, and Stephanopoulos G. Interactive exploration of microarray gene expression patterns in a reduced dimensional space. Genome Res 12: 1112–1120, 2002.[Abstract/Free Full Text]
  20. Nadon R and Shoemaker J. Statistical issues with microarrays: processing and analysis. Trends Genet 18: 265–271, 2002.[CrossRef][ISI][Medline]
  21. Richer JK, Jacobsen BM, Manning NG, Abel MG, Wolf DM, and Horwitz KB. Differential gene regulation by the two progesterone receptor isoforms in human breast cancer cells. J Biol Chem 277: 5209–5218, 2002.[Abstract/Free Full Text]
  22. Saito-Hisaminato A, Katagiri T, Kakiuchi S, Nakamura T, Tsunoda T, and Nakamura Y. Genome-wide profiling of gene expression in 29 normal human tissues with a cDNA microarray. DNA Res 9: 35–45, 2002.[Abstract]
  23. Schug J, Schuller W, Kappen C, Salbaum M, Bucan M, and Stoeckert CJ Jr. Promoter features related to tissue specificity as measured by Shannon entropy. Genome Res 6: R33.31–R33.24, 2005.
  24. Shyamsundar R, Kim YH, Higgins JP, Montgomery K, Jorden M, Sethuraman A, van de Rijn M, Botstein D, Brown PO, and Pollack JR. A DNA microarray survey of gene expression in normal human tissues. Genome Biol 6: R22, 2005.[CrossRef][Medline]
  25. Singh A and Nocerino J. Robust intervals for some environmental applications. Chemometrics and Intelligent Laboratory Systems 37: 55–69, 1997.
  26. Su AI, Wiltshire T, Batalov S, Lapp H, Ching KA, Block D, Zhang J, Soden R, Hayakawa M, Kreiman G, Cooke MP, Walker JR, and Hogenesch JB. A gene atlas of the mouse and human protein-encoding transcriptomes. Proc Natl Acad Sci USA 101: 6062–6067, 2004.[Abstract/Free Full Text]
  27. 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, and Friend SH. Gene expression profiling predicts clinical outcome of breast cancer. Nature 415: 530–536, 2002.[CrossRef][Medline]
  28. Venter JC, Adams MD, Myers EW, Li PW, Mural RJ, Sutton GG, Smith HO, Yandell M, Evans CA, Holt RA, Gocayne JD, Amanatides P, Ballew RM, Huson DH, Wortman JR, Zhang Q, Kodira CD, Zheng XH, Chen L, Skupski M, Subramanian G, Thomas PD, Zhang J, Gabor Miklos GL, Nelson C, Broder S, Clark AG, Nadeau J, McKusick VA, Zinder N, Levine AJ, Roberts RJ, Simon M, Slayman C, Hunkapiller M, Bolanos R, Delcher A, Dew I, Fasulo D, Flanigan M, Florea L, Halpern A, Hannenhalli S, Kravitz S, Levy S, Mobarry C, Reinert K, Remington K, Abu-Threideh J, Beasley E, Biddick K, Bonazzi V, Brandon R, Cargill M, Chandramouliswaran I, Charlab R, Chaturvedi K, Deng Z, Di Francesco V, Dunn P, Eilbeck K, Evangelista C, Gabrielian AE, Gan W, Ge W, Gong F, Gu Z, Guan P, Heiman TJ, Higgins ME, Ji RR, Ke Z, Ketchum KA, Lai Z, Lei Y, Li Z, Li J, Liang Y, Lin X, Lu F, Merkulov GV, Milshina N, Moore HM, Naik AK, Narayan VA, Neelam B, Nusskern D, Rusch DB, Salzberg S, Shao W, Shue B, Sun J, Wang Z, Wang A, Wang X, Wang J, Wei M, Wides R, Xiao C, Yan C, et al. The sequence of the human genome. Science 291: 1304–1351, 2001.[Abstract/Free Full Text]
  29. Walker MG. Pharmaceutical target identification by gene expression analysis. Mini Rev Med Chem 1: 197–205, 2001.[CrossRef][Medline]
  30. Yandell BS. Practical Data Analysis for Designed Experiments. London: Chapman & Hall, 1997.
  31. Zar JH. Biostatistical Analysis. Upper Saddle River, NJ: Prentice Hall, 1999.




This Article
Free upon publication Free Article
Right arrow Abstract Freely available
Right arrow Full Text (PDF)
Right arrow Supplementary Material
Right arrow All Versions of this Article:
26/2/158    most recent
00313.2005v1
Right arrow Alert me when this article is cited
Right arrow Alert me if a correction is posted
Right arrow Citation Map
Services
Right arrow Email this article to a friend
Right arrow Similar articles in this journal
Right arrow Similar articles in PubMed
Right arrow Alert me to new issues of the journal
Right arrow Download to citation manager
Citing Articles
Right arrow Citing Articles via ISI Web of Science (3)
Right arrow Citing Articles via Google Scholar
Google Scholar
Right arrow Articles by Liang, S.
Right arrow Articles by Liu, W.
Right arrow Search for Related Content
PubMed
Right arrow PubMed Citation
Right arrow Articles by Liang, S.
Right arrow Articles by Liu, W.

HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS
Visit Other APS Journals Online
Copyright © 2006 by the American Physiological Society.