Bayesian integrated modeling of expression data: a case study on RhoG
© Gupta et al; licensee BioMed Central Ltd. 2010
Received: 4 June 2009
Accepted: 1 June 2010
Published: 1 June 2010
DNA microarrays provide an efficient method for measuring activity of genes in parallel and even covering all the known transcripts of an organism on a single array. This has to be balanced against that analyzing data emerging from microarrays involves several consecutive steps, and each of them is a potential source of errors. Errors tend to accumulate when moving from the lower level towards the higher level analyses because of the sequential nature. Eliminating such errors does not seem feasible without completely changing the technologies, but one should nevertheless try to meet the goal of being able to realistically assess degree of the uncertainties that are involved when drawing the final conclusions from such analyses.
We present a Bayesian hierarchical model for finding differentially expressed genes between two experimental conditions, proposing an integrated statistical approach where correcting signal saturation, systematic array effects, dye effects, and finding differentially expressed genes, are all modeled jointly. The integration allows all these components, and also the associated errors, to be considered simultaneously. The inference is based on full posterior distribution of gene expression indices and on quantities derived from them rather than on point estimates. The model was applied and tested on two different datasets.
The method presents a way of integrating various steps of microarray analysis into a single joint analysis, and thereby enables extracting information on differential expression in a manner, which properly accounts for various sources of potential error in the process.
Microarrays are popular high-throughput biological assays that measure the expression level of thousands of genes in the biological samples and generate large, complex datasets. In spite of the advances in technology, it is a major challenge to produce reliable gene expression data with a high signal-to-noise ratio, and analyze these large datasets in an adequate manner. Analyzing microarray data is usually performed in a step-wise manner, starting with, (i) normalization of the intensity measurements, to adjust or account for systematic technical variation, (ii) correcting dye-bias if dye-bias remains after normalization, (iii) identifying differentially expressed genes on the normalized data, and completing the analysis with (iv) functional annotation of the differentially expressed genes. All these steps are regarded as independent, but they are crucial for any biologically meaningful analysis.
Normalization is an integral part of the analysis, aiming at retaining the systematic effects resulting from the biological process of interest while removing the systematic technical variations occurring due to experimental variability. Normalization has researched for quite some time and publications proposing new procedures are available [1–4]. Some datasets display a consistent bias for a given probe in either Cy3 or Cy5 direction even after the data have been normalized using median-centered and lowess normalization methods. This bias is called dye bias and it is observed on a variety of platforms and labeling systems, including PCR-spotted and short oligonucleotide labeling methods. Many experimentalists and statisticians recommend using a dye-swap design to correct for this bias. Some publications have shown by considering experimental data that, if uncorrected, this bias can lead to the erroneous identification of genes [5–7].
Identification of differentially expressed genes is usually the main goal of microarray experiment. Chen et al.  assessed differentially expressed genes by calculating fold changes between genes under different conditions. Fold-change method, the simplest and the most intuitive method for finding genes that are differentially expressed, has many drawbacks. Later, improved methods based on t-test, regularized t-test [9, 10] were proposed. Model based approaches have also been published to identify differentially expressed genes. Most methods listed in the literature use point estimates of expression and depend upon replicates available for the estimation of variances.
Step-wise analysis of the microarray data has two major drawbacks: (i) output from one step acts as direct input to the next, without attempting to account for the uncertainties associated with the value that was obtained; as a consequence, (ii) re-analyzing the data by altering the method used for a single step will often produce conflicting results. For this reason, Bhattacharjee et al.  proposed a method that aims at integrating the independent steps, so that uncertainties from each step could be accounted systematically. Lewin et al.  also proposed an integration of the normalization and classification step by using a Hierarchical Bayesian model. These proposed integrated approaches performed better than their step-wise approach counterparts. Moreover, the Bayesian formulation enables a much richer output than current step-wise analyses.
In here, we also propose an integrated statistical model under the Bayesian framework, where normalization and differential expression are modeled jointly, and correction of the saturated signal is also incorporated. Saturation refers to the optical saturation and not chemical saturation. Such (optical) signal saturation occurs in the scanning of hybridized arrays when the digitalized signal from a pixel exceeds the scanner's upper threshold of detection (216-1 = 65535, for a 16 bit computer storage system). Saturation causes a downward bias in gene expression measurements, which then affects high level analysis, such as class prediction, class comparison or clustering that utilizes these signals .
Usually, data extracted from a single scan and a single scanner setting is used for all high level analyses. However, a single setting is unable to capture correctly the expression of both weakly and highly expressed genes. As a result, the sensitivity level of the scanner is adjusted to get reliable measurements from all fluorescent spots present on the hybridized array. Scanner sensitivity has to be raised to a certain level to ensure that the signal from weakly expressed genes exceeds the intrinsic noise level of the scanner, but this causes saturation for highly expressed genes. Several methods [14–19] have been proposed for correcting the bias caused by signal saturation.
In here, we extend our previous work (Gupta et al. ) on handling signal saturation by using several scans at varying scanner sensitivities. We propose an integrated statistical approach where correcting signal saturation, systematic array effects, gene-specific dye effects, and differential expression are modeled simultaneously. We estimate our model in a fully Bayesian way with the WinBUGS software . The Bayesian framework allows for joint estimation of a large number of parameters, and enables us to obtain here the posterior distribution of any parameter in the model and of any function of such parameters. We show how to exploit these posterior distributions to assess differential expression, using multiple criteria for this purpose. The uncertainties in the parameter estimates are thereby incorporated in a natural manner into a proposed list of candidate genes.
RhoG is a protein belonging to the family of the small GTPases [21, 22]. It is involved in several intracellular signaling pathways regulating cell motility and adhesion to extracellular matrix. Together with Cdc42 and Rac1, RhoG is able to elicit formation of both filopodia and lamellipodia. Neurite formation and regulation of axon dynamics in neurons are more specific functions in which RhoG is acting together with other Rho proteins and their interactors. Within the cells, Rho proteins can be found in an active form and inactive form. Mutants of RhoG (RhoG12 and RhoG17) can be used to keep the protein in a constitutively activated (mutation of the 12th amino acid) or inactivated (mutation of the 17th amino acid) form. In this study we investigate effect of mutants RhoG12 and RhoG17 on the gene expression of HeLa cell lines.
Design details along with the configuration of PMT used to obtain multiple scans for two replicate arrays of dataset-1.
Design details along with the combinations of PMT and LP used to obtain multiple scans for three replicate arrays of dataset-2.
For details about RNA extraction, probes labeling, and microarray hybridization for the two datasets, see Additional file-3.
Bayesian hierarchical model
The model aims at finding differentially expressed genes under c max conditions (here c max = 2, experimental and control, but c max can be more than two, for example, when comparing multiple conditions over time), each replicated on r max arrays (here r max = 2 (for dataset-1) and 3(for dataset-2)), and each array scanned s times (here s max = 3) under different scanner settings. We assume that, under condition c, each gene i has an underlying signal, which cannot be measured directly. We call this signal the true latent intensity of the gene under condition c and denote it by T ic , c = 1, 2; i = 1, 2, ..., N, where N is the number of spots used in the experiment. The entire model is defined on the logarithmic scale, base e.
Signal correction is done separately for each replicate by combining three scans made by varying the scanner settings for that replicate. Let Q icr represent latent intensity of gene i under condition c on replicate r. The scanner settings used in the first scan for each replicate are chosen to correspond to the situation, where only a single scan would be made; therefore these first scans form a natural basis for calibrating the latent intensities Q icr . They are also expected to capture, without a downward bias caused by saturation, spots that do not have abundant levels of RNA. The second and the third scans were made by choosing the scanner settings so that their measured signals would be weaker. Latent intensities corresponding to the second and third scans are now assumed to be linked to Q icr by simple functional relationships, respectively by fcr 2(Q icr ) and fcr 3(Q icr ) (discussed briefly later).
where, T ic is the true latent intensity of a gene i under condition c, A ir is the array effect, and β i is the gene-specific dye effect. Since for cDNA experiments both the control and the experimental samples are hybridized on the same array, the array effect (A ir ) is not dependent on the condition c. The gene-specific dye-bias correction (β i ) is only applied when the values are taken from Cy5 intensity data, as enforced by the indicator function I(Cy5) cr. However, the symmetric model in which the correction is applied to Cy3 channel only would perform identically with the difference that the bias terms would be negated. A similar gene-specific dye bias correction was used in Kelley et al. .
where, L(I k ) is the length of the k th interval. The array effects (A ir ) are estimated over the set of intervals I1, I2, ... I k , subject to the constraints ∑ r A jr = 0, j = 1, 2, ...., k to ensure identifiability. Estimation of array effects over a set of intervals is similar to the intensity based estimation of array effects previously reported in Yang et al.  and Dudoit et al. .
To complete the specification of the model, we assumed Uniform prior distribution over the interval [0, 15] on logarithmic scale for T ic . The array effects A jr were assigned Normal priors with mean 0 and precision 0.1 (inverse of variance). The parameters b jcr and d jcr were assigned Uniform priors over the interval [0, 5]. Gene specific dye effects β i were also assigned Normal priors with mean 0 and precision 0.1. The errors ε icrs are assumed to be independent and identically distributed Normal random variables with mean 0 and interval dependent variances η2 jcrs , where s = 1, 2, 3; j = 1, 2, ...., k. The interval dependent precision parameters (η jcr1 2, η jcr2 2, and η jcr3 2; j = 1, 2, ..., k) were assigned gamma priors with parameters (0.001, 0.001).
Finally, as per Gupta et al. , to account for the effect of saturation, we treated signal measurements exceeding the threshold of loge(45000) as 'missing data'. We compensated for the resulting loss of information by applying model-based data augmentation and using the measurements taken from the second and/or the third scan which had been obtained by varying scanner settings.
The model was formulated in BUGS language and parameter estimation was performed using WinBUGS .
Rules for selecting genes
Genes are selected as being potentially up-regulated if p i + > p cut and down-regulated if p i - > p cut , where again the cut-off point p cut needs to be chosen in advance. These posterior probabilities (p i + and p i -) are easily estimated by counting the proportion of MCMC samples in which the chosen criteria are satisfied. The choice of the controlling threshold values p cut , D thres + and D thres - depends on the biological question being studied, and can be problematic to choose. However, in practice, the values are chosen only after a preliminary analysis of the data.
The above-mentioned criterion is quite similar to the criterion used in Lewin et al. , for selecting interesting genes. Other criteria for ranking genes include the use of standardized differences, z i = mean(D i )/sd(D i ), and determining the highest percentile for which the credibility interval for D i does not cover zero . It is important to note that identification of differentially expressed genes is here based directly on determining the gene-wise posterior probabilities that the latent 'true' difference in expression in the two conditions exceeds a certain threshold. Thus our method does not use the general framework of statistical hypothesis testing, involving, for example, p-values, or corrections of significance levels to account for multiple testing. Unlike Lewin et al. , we also have here not made an attempt to calibrate the chosen thresholds on the basis of frequentist criteria such as False Discovery/Non-Discovery Rate.
Results and Discussion
Application to dataset-1
The model under "Bayesian hierarchical model" without parameter (β i ) was applied to dataset-1 to illustrate the criterion presented under "Rules for selecting genes". Since both replicate arrays from dataset-1 have the same dye-orientation, the dye-bias in the data cannot be assessed.
Computational details and parameter estimation
For dataset-1, foreground median values for each condition without background correction were used for the analysis. As a result, we had no negative values. This particular dataset had 43,376 genes (on single array) × 2 (replicates used) × 3 (scans used) × 2 (dyes/conditions) = 520,512 data points to be used for the analysis. The current model runs in OpenBUGS version 2.01 on Intel Pentium processor 2.80 GHz with 1 GB RAM and takes approximately 4 seconds per iteration using two chains in parallel. Convergence was monitored visually (i.e. by the mixing of two chains) and two chains of 10,000 iterations each were generated to check the convergence of the parameter estimates under consideration. Thereafter a sample of size 10,000 was generated to make inference.
Posterior median estimates of the parameters (b, d) for two conditions over replicate-1 of dataset-1.
Posterior median estimate of b, d(median ± sd) for the two condition over replicate-1
0.9158 ± 0.0001
0.8370 ± 0.0001
0.9009 ± 0.0001
0.7952 ± 0.0001
0.9084 ± 0.0003
0.8275 ± 0.0003
0.9230 ± 0.0003
0.8325 ± 0.0003
0.9116 ± 0.0005
0.8354 ± 0.0005
0.9344 ± 0.0005
0.8554 ± 0.0005
0.9206 ± 0.0006
0.8554 ± 0.0006
0.9436 ± 0.0006
0.8798 ± 0.0006
Posterior median estimates of the array effect over the four intervals and over two replicates of dataset-1.
Posterior median estimate of array effect (median ± sd) over replicates
0.0018 ± 0.0006
-0.0094 ± 0.0006
-0.3107 ± 0.0039
0.3143 ± 0.0039
-0.3288 ± 0.0061
0.3302 ± 0.0061
-0.2883 ± 0.0049
0.2910 ± 0.0049
The breakpoints were selected using visual inspection, but it would also be possible to treat them as model parameters and then estimate them jointly with b jcr , d jcr and A jr . This was not done here because of the additional computational burden that would have resulted in analyzing the huge dataset.
Discussion of decision rules
Brief description and comments on some genes (of datset-1) found to be differentially expressed and associated with RhoG from literature.
Fold change(natural scale)
Pubmed Id (PMID)
ARHGEF3 form complex with G proteins and stimulate Rho-dependent signals.
p+ = 1
ICAM1 binds to integrins of type CD11a/CD18, or CD11b/CD18 and stimulates intercellular signaling.
p+ = 0.9913
IL6 is an immunoregulatory cytokine that activates a cell surface signaling assembly composed of IL6, IL6RA, and the shared signaling receptor gp130.
p+ = 1
This gene encodes a protein which interacts directly with specific target DNA sequences to regulate gene expression.
12739001, 1620121, 9671479, 10744696
p+ = 0.9935
NFKB is a transcription regulator that is activated by various intra-and extra-cellular stimuli. Activated NFKB translocates into the nucleus and stimulates the expression of genes involved in a wide variety of biological functions.
12670394, 11803464, 12376551
p+ = 0.9942
NISCH is involved in the regulation of cell migration and cell invasion.
p+ = 0.9965
PCNA is found in the nucleus and is a cofactor of DNA polymerase delta. The encoded protein helps increase the processivity of leading strand synthesis during DNA replication.
p- = 1
Prostaglandin-endoperoxide synthase is the key enzyme in prostaglandin biosynthesis, and acts both as a dioxygenase and as a peroxidase.
p+ = 1
RHOF functions cooperatively with CDC42 and Rac to generate filopodia increasing the diversity of actin-based morphology.
p+ = 0.9994
Gene ontology categories enriched among the differentially expressed genes
Some selected GO categories, along with the numbers of varying and analyzed genes from dataset-1.
Gene ontology categories
Number of genes estimated as varying
Number of genes analyzed
Cellular component organization and biogenesis
Organelle organization and biogenesis
Establishment and/or maintainence of chromatin architecture
Regulation of actin cytoskeleton dynamics is one of the central effects of RhoG on cells. RhoF (or Rif) was one of the genes that showed up in this category . RhoF is involved in the filopodia formation through mDia2. Among the small GTPases, RhoA is a key regulator of actin cytoskeleton. Presently, little is known about the possible functional relationships of RhoA and RhoG. However, we identified several interesting candidate genes that could participate in the possible cross-talk between these Rho proteins: ROCK2 is a classical RhoA-linked regulator of actin  and two RhoA GEFs (ARHGEF10L  and ARHGEF3 ) exhibit ways in which RhoG could regulate the activity of RhoA by inducing the expression of their regulators.
We also identified Cdc42 regulators (Chiamerin, see Additional file 1, qPCR results) indicating that there are unknown cross-talk between RhoG and other RhoGTPases in regulating actin cytoskeleton homeostasis. Moreover, ARPC3, a part of the Arp2/3 complex, was identified [29, 30]. This complex is one of the actin nucleation apparatuses responsible for many actin-related functions like endosytosis, lamellipodia formation and filopodia formation. Our list of candidate genes helps us understand how regulatory genes like RhoG are performing their multitasking in cell dynamics.
Step-wise analysis using existing approaches
For a comparison, dataset-1 was also analyzed in a step-wise manner using the existing popular softwares/procedures. The data from the multiple scans of each replicate and from the two dyes were first combined using the multiscan package in R. The multiscan package implements the method of Khondoker et al. , for estimating gene expressions from multiple laser scans of hybridized microarrays. The method proposed in Khondoker et al.  has already been compared with a similar method from Gupta et al.  which was utilized in this paper for estimating gene signals from multiple scans. Gupta et al.  also showed that the estimated gene signal from multiple scans gave better results when utilized for high level analysis than the gene signal data from a single scan.
The combined signals from the multiple scans of the three replicates and for the two dyes were normalized using Quantile normalization method in R . Limma was used to fit a model and to identify differentially expressed genes. We used DAVID  for the functional annotation of the selected genes. This step-wise analysis identified three broad functionalities "cell differentiation", "cell cycle" and "developmental process" (also listed in Table 6, results from integrated approach) but failed to identify other specific functionalities associated with the experiment.
Assessing dye bias
Availability and requirements
Project name: Bayesian Integrated analysis
Operating system(s): Platform independent
Programming language: WinBUGS
License: Code is freely available for usage and modifications; however, appropriate reference of this article is essential.
Our focus has been on modeling differential gene expression between two experimental conditions, by proposing an integrated statistical solution where signal correction, systematic array and dye effects, and differential expression, were all modeled jointly. All processing steps were integrated into a common statistically coherent framework, allowing all components and their associated errors to be considered simultaneously. The inference was based on the full posterior distribution of gene expression indices and of their derived quantities, such as difference (D i ), rather than on point estimates. In this respect, our approach differs in a fundamental way from most alternative methods which have been proposed in the literature and are build on the idea of statistical significance testing.
The key advantages of the proposed integrated analysis are: (i) robustness of final results towards small variations in outcomes of intermediate steps of the analysis, and (ii) straightforward interpretability of results, when stated in terms of the posterior distributions of differences between the true expression levels obtained under different experimental conditions.
The Bayesian hierarchical models considered here are a step towards a complete integrated approach to the analysis of gene expression data. In future, the model presented here can be extended to include other common steps in the analysis, such as background correction, quality inspection, functional annotation, and clustering. Simultaneous consideration of such additional steps can be expected to lead to further improvements in the estimates and thus to more reliable inferences.
The current model was successfully implemented using WinBUGS software. WinBUGS provides a user-friendly and easily modifiable implementation of Bayesian hierarchical models. This ease of handling and modifying complicated models is balanced by the running time when dealing with large genomic application data. All future extensions (say, incorporating background correction) need to be implemented in C or C++ for a realistic running time of the models. However, comparison of multiple conditions using the integrated model in BUGS (described in here) can be easily speeded up by running the same model with different conditions on different machines.
The results we have obtained from the RhoG experiments are very interesting and provide us several interesting candidate genes for further studies. Many of the genes identified suggest novel links with the cellular machinery.
The authors thank Andrew Thomas for his help and comments during implementation and Panu Somervuo for helping us with the preliminary analysis. We also thank Eeva-Marja Turkki for running the qPCR analysis. This study was supported by the Maj and Tor Nessling Foundation.
- Yang YH, Dudoit S, Luu P, Lin DM, Peng V, Ngai J, Speed T: Normalization for cDNA microarray data: a robust composite method addressing single and multiple slide systematic variation. Nucleic Acid Res 2002, 30: E15. 10.1093/nar/30.4.e15View ArticlePubMedPubMed CentralGoogle Scholar
- Tseng GC, Oh M-K, Rohlin L, Liao JC, Wong WH: Issues in cDNA microarray analysis: quality filtering, channel normalization, models of variations and assessment of gene effects. Nucleic Acids Res 2001, 29: 2549–2557. 10.1093/nar/29.12.2549View ArticlePubMedPubMed CentralGoogle Scholar
- Workman C, Jensen LJ, Jarmer H, Berka R, Gautier L, Nielsen HB, Saxild H-H, Nielsen C, Brunak S, Knudsen S: A new non-linear normalization method for reducing variability in DNA microarray experiments. Genome Biol 2002, 3(9):research0048. 10.1186/gb-2002-3-9-research0048View ArticlePubMedPubMed CentralGoogle Scholar
- Dudoit S, Yang YH, Luu P, Speed TP: Normalization for cDNA microarray data. In Microarrays:Optical Technolologies and Informatics, Vol. 4266 of Proceedings of SPIE Edited by: Bittner ML, Chen Y, Dorsel AN, Dougherty ER. 2001: 141–152.Google Scholar
- Rosenzweig BA, Pine PS, Domon OE, Morris SM, Chen JJ, Sistare FD: Dye bias correction in dual-labeled cDNA microarray gene expression measurements. Environ Health Perspect 2004, 112(4):480–487.View ArticlePubMedPubMed CentralGoogle Scholar
- Martin-Magniette M-L, Aubert J, Cabannes E, Daudin J-J: Evaluation of the gene-specific dye bias in cDNA microarray experiments. Bioinformatics 2005, 21(9):1995–2000. 10.1093/bioinformatics/bti302View ArticlePubMedGoogle Scholar
- Kelley R, Feizi H, Ideker T: Correcting for gene-specific dye bias in DNA microarrays using the method of maximum likelihood. Bioinformatics 2007, 24: 71–77. 10.1093/bioinformatics/btm347View ArticlePubMedPubMed CentralGoogle Scholar
- Chen Y, Dougherty ER, Bittner ML: Ratio-based decisions and the quantitative analysis of cDNA microarray images. J Biomed Opt 1997, 2: 363–374. 10.1117/12.281504View ArticleGoogle Scholar
- Baldi P, Long AD: A Bayesian framework for the analysis of microarray expression data: regularized t-test and statistical inferences of gene changes. Bioinformatics 2001, 17: 509–519. 10.1093/bioinformatics/17.6.509View ArticlePubMedGoogle Scholar
- Tusher VG, Tibshirani R, Chu G: Significance analysis of microarrays applied to the ionizing radiation response. Proc Natl Acad Sci 2001, 98: 5116–5121. 10.1073/pnas.091062498View ArticlePubMedPubMed CentralGoogle Scholar
- Bhattacharjee M, Pritchard CC, Nelson PS, Arjas E: Bayesian integrated functional analysis of microarray data. Bioinformatics 2004, 20: 2943–2953. 10.1093/bioinformatics/bth338View ArticlePubMedGoogle Scholar
- Lewin A, Richardson S, Marshall C, Glazier A, Aitman T: Bayesian modeling of differential gene expression. Biometrics 2006, 62: 1–9. 10.1111/j.1541-0420.2005.00394.xView ArticlePubMedGoogle Scholar
- Hsiao L, Jenser R, Yoshida T, Clark K, Blumenstock J, Gullans S: Correcting for signal saturation errors in the analysis of microarray data. Biotechniques 2002, 32: 330–336.PubMedGoogle Scholar
- Lyng H, Badiee A, Svendsrud DH, Hovig E, Myklebost O, Stokke T: Profound influence of microarray scanner characteristics on gene expression ratios: analysis and procedure for correction. BMC Genomics 2004, 5: 10. 10.1186/1471-2164-5-10View ArticlePubMedPubMed CentralGoogle Scholar
- Piepho HP, Keller B, Hoecker N, Hochholdinger F: Combining signals from spotted cDNA microarrays obtained at different scanning intensities. Bioinformatics 2006, 22: 802–807. 10.1093/bioinformatics/btk047View ArticlePubMedGoogle Scholar
- Skibbe DS, Wang X, Zhao X, Borsuk LA, Nettleton D, Schnable PS: Scanning microarrays at multiple intensities enhances discovery of differentially expressed genes. Bioinformatics 2006, 22: 1863–1870. 10.1093/bioinformatics/btl270View ArticlePubMedGoogle Scholar
- Khondoker MR, Glasbey CA, Worton BJ: Statistical estimation of gene expression using multiple laser scans of microarrays. Bioinformatics 2006, 22: 215–219. 10.1093/bioinformatics/bti790View ArticlePubMedGoogle Scholar
- Gupta R, Auvinen P, Thomas A, Arjas E: Bayesian hierarchical model for correcting signal saturation in microarrays using pixel intensities. Statistical Application in Genetics and Molecular Biology 2006., 5: Article 20. Article 20.Google Scholar
- Gupta R, Arjas E, Kulathinal S, Thomas A, Auvinen P: Bayesian hierarchical model for estimating gene expression intensity using multiple scanned microarrys. EURASIP Journal on Bioinformatics and Systems Biology 2008. Article ID 231950. Article ID 231950.Google Scholar
- Spiegelhalter DJ, Thomas A, Best NG: WinBUGS, Version 1.2. User Manual, MRC Biostatistics Unit; 1999.Google Scholar
- Gauthier-Rouvière C, Vignal E, Mériane M, Roux P, Montcourier P, Fort P: RhoG GTPase controls a pathway that independently activates Rac1 and Cdc42Hs. Mol Biol Cell 1998, 9: 1379–1394.View ArticlePubMedPubMed CentralGoogle Scholar
- Govek E-E, Newey SE, Aelst LV: The role of the Rho GTPases in neuronal development. Genes & Dev 2005, 19: 1–49. 10.1101/gad.1256405View ArticleGoogle Scholar
- Hein A-MK, Richardson S, Causton HC, Ambler GK, Green PJ: BGX: A fully Bayesian gene expression index for Affymetrix GeneChip data. Biostatistics 2005, 6(3):349–373. 10.1093/biostatistics/kxi016View ArticlePubMedGoogle Scholar
- Dennis G Jr, Sherman BT, Hosack DA, Yang J, Gao W, Lane HC, Lempicki RA: DAVID: Database for annotation, visualization, and integrated discovery. Genome Biol 2003, 4(5):P3. 10.1186/gb-2003-4-5-p3View ArticlePubMedGoogle Scholar
- Pellegrin S, Mellor H: The Rho family GTPase Rif induces filopodia through mDia2. Curr Biol 2005, 15: 129–133. 10.1016/j.cub.2005.01.011View ArticlePubMedGoogle Scholar
- Nakagawa O, Fujisawa K, Ishizaki T, Saito Y, Nakao K, Narumiya S: ROCK-I and ROCK-II, two isoforms of Rho-associated coiled-coil forming protein serine/threonine kinase in mice. FEBS Lett 1996, 392: 189–193. 10.1016/0014-5793(96)00811-3View ArticlePubMedGoogle Scholar
- Winkler S, Mohl M, Wieland T, Lutz S: GrinchGEF--A novel Rho-specific guanine nucleotide exchange factor. Biochemical and Biophysical Research Communications 2005, 335: 1280–1286. 10.1016/j.bbrc.2005.08.025View ArticlePubMedGoogle Scholar
- Arthur WT, Ellerbroek SM, Der CJ, Burridge K, Wennerberg K: XPLN, a guanine nucleotide exchange factor for RhoA and RhoB, But not RhoC. J Biol Chem 2002, 277: 42964–42972. 10.1074/jbc.M207401200View ArticlePubMedGoogle Scholar
- Welch MD, DePace AH, Verma S, Iwamatsu A, Mitchison TJ: The Human Arp2/3 complex is composed of evolutionarily conserved subunits and is localized to cellular regions of dynamic actin filament assembly. J Cell Biol 1997, 138: 375–384. 10.1083/jcb.138.2.375View ArticlePubMedPubMed CentralGoogle Scholar
- Zerial M, McBride H: Rab proteins as membrane organizers. Nat Rev Mol Cell Biol 2001, 2(2):107–17. 10.1038/35052055View ArticlePubMedGoogle Scholar
- Bolstad BM, Irizarry RA, Astrand M, Speed TP: A Comparison of Normalization Methods for High Density Oligonucleotide Array Data Based on Bias and Variance. Bioinformatics 2003, 19(2):185–193. 10.1093/bioinformatics/19.2.185View ArticlePubMedGoogle Scholar
This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.