REST 2008 software - Greater certainty in expression studies

The new stand-alone software versions REST 2008 was programmed and designed by Matthew Herrmann, David Chiew, and Birgit Speller working at Corbett Research (Sydney, Australia) and Michael W. Pfaffl, Technical University of Munich, Germany.
REST 2008 builds on its predecessor REST 2005 with significant improvements to randomization algorithms. This new revision introduces alternative data inputs such as single run efficiency and amplification take-off point, alleviating the need to set amplification plot thresholds.

Download => Manual REST 2008

Download => REST 2008 software (removed) Slide show REST-2008

Please download newest version REST 2009 => http://www.REST.de.com

Abstract
REST 2008 is a standalone software package for analyzing gene expression using real-time amplification data. The software addresses issues surrounding the measurement of uncertainty in expression ratios by introducing randomization and bootstrapping techniques. New confidence intervals for expression levels also allow measurement of not only the statistical significance of deviations but also their likely magnitude, even in the presence of outliers. Whisker box plots provide a visual representation of variation for each gene, highlighting potential issues such as distribution skew. REST 2008 builds on its predecessor REST 2005 with significant improvements to randomization algorithms. This new revision introduces alternative data inputs such as single sample efficiency and amplification take-off point, alleviating the need to set amplification plot thresholds.

What's NEW since REST 2005 ?

NEW - REST-RG mode

A new method of input has been introduced, allowing users to copy and paste results from the Rotor-Gene software's Comparative Quantitation analysis. This is an alternative to importing standard curve and CT results. See REST-RG Mode chapter for more details.

NEW - Whisker-Box Plots Exportable

Whisker-Box plots can now be exported by right-clicking on the graph.

NEW - Improved randomisation

Improvements to the randomisation algorithms have been made, making confidence intervals much tighter, and p-values more accurate. In previous versions the pair-wise fixed reallocation was incorrectly matching the gene of interest CT with the incorrect reference CT, this issue has been rectified in REST 2008.

NEW - Handling of standard curve variation

REST 2008 no longer takes into account the variation of the standard curve and implements improvements to the calculation of confidence intervals and p-values. In previous versions, the software would randomly pick two points from the standard curve and calculate an efficiency based on that. However there is a situation when two points are chosen that lie close to each other on the standard curve, this can cause a bogus efficiency which adds unnecessary outliers to the random distribution. We now calculate the efficiency by determining the line of best fit for the standard curve, this efficiency is used through the randomisation process.

Why REST

Prior to REST (Relative Expression Software Tool, Pfaffl et al 2002), Relative Quantitation in qRT-PCR was a technique which allowed the estimation of gene expression. While useful, it did not provide statistical information suitable for comparing groups of treated versus untreated samples in a robust fashion. To illustrate with an example, let us say we are testing to see if a particular mRNA is responsible for sending pain messages. We split up our patients into two groups: one which will be subject to pain (such immersion of the hand in ice-cold water), and the other, which is our control group. Following this, we measure the quantities of gene of interest mRNA in both groups, relative to reference genes. Our question is: did the group subject to pain release more gene of interest mRNA than the other? Prior approaches are insufficient to answer this question. They may calculate an average expression value indicating whether a particular subject in one group appeared to release more or less gene of interest mRNA than another subject, but without any statistical test to determine accuracy. Due to the use of ratios in gene expression, it becomes very complex to perform traditional statistical analysis, as ratio distributions do not have a standard deviation. REST 2005 overcomes these problems by using simple statistical randomisation tests. Such tests can appear counter-intuitive and so it is recommended to read the discussions on randomisation techniques in the topic Links before continuing.

Hypothesis Test

The purpose of REST 2008 is to determine whether there is a significant difference between samples and controls, while taking into account issues of reaction efficiency and reference gene normalisation. Because the normalisation and efficiency calculations involve ratios and multiple sources of error, it would be extremely difficult to devise a traditional statistical test, and so randomisation techniques are used instead.

The hypothesis test P(H1) indicated in the results table, represents the probability of the alternate hypothesis that the difference between sample and control groups is due only to chance. To devise a strong randomisation test, we use the following randomisation scenario: "If any perceived variation between samples and controls is due only to chance, then we could randomly swap values between the two groups and not see a greater difference than what we see between the labelled groups."

The hypothesis test performs a large number of random reallocations of samples and controls between the groups. It then counts the number of times the relative expression of the randomly assigned group is greater than the sample data.

Reference Gene Normalisation

REST 2008 allows the researcher to take into consideration multiple reference genes when determining expression, although it still remains possible to use a single reference. When estimating a sample's expression ratio, an intermediate absolute concentration value is calculated according to the following formula:

concentration = efficiencyavg(Controls) – avg(Samples)

Errors in calculation of concentration occur due to linear variation in CT values. Estimates of concentration use an equation of the form c = A x e^Ct and so vary exponentially.

This formula is used to obtain mean estimates of the uncorrected absolute concentration for each gene. For a single reference gene, the concentration of the gene of interest is divided by the reference gene value to obtain an expression level, as is done in the Two Standard Curve technique:

expression = goiConcentration ÷ refConcentration

For multiple reference genes, the geometric mean is taken of all reference gene concentrations, since concentration estimates vary exponentially (Vandesompele et al., 2002):

expression = goiConcentration ÷ GEOMEAN (refConc1, refConc2,, …)

Alternatively, to normalise according to multiple reference genes, a second approach can be used, to normalise the individual expressions relative to each reference gene which represents an alternative approximations of the true expression value. To take all into account simultaneously, they are averaged using a geometric mean (since ratios are being used):

expression = GEOMEAN (goiConcentration ÷ refConc1, goiConcentration ÷ refConc2, …)

Since the mean concentrations of each gene do not change, they can be calculated at the beginning of the algorithm, and expressed as a single value, called the "Normalisation factor", equal to their geometric mean.

Greater Accuracy for Hypothesis Tests

The redevelopment of the REST 2005 software as a stand-alone application provides an order of magnitude of increase in performance. The speed improvements have been used to increase the number of randomisation iterations from 2,000 to 50,000, compared to earlier REST versions (Pfaffl et al., NAR 2002), increasing the accuracy and reproducibility of hypothesis tests to a level equivalent to traditional statistical tests.

Expression Level Confidence Intervals

While previous REST publications provide a means of determining the mean output and a P value for the likelihood of up or down-regulation using a hypothesis test, bootstrapping techniques can be used to provide 95% confidence intervals for expression ratios, without normality or symmetrical distribution assumptions. While a hypothesis test provides a measure of whether there was a statistically significant result, the confidence interval provides a range that can be checked for semantic significance. For example, drinking cough medicine before driving may increase the chances of an accident by 1x10^-6 %. While a statistical test may show the difference to be significant, it clearly poses no real threat to drivers, when taking into consideration the average number of accidents a driver has in their lifetime.

Efficiency Error Measurement

All statistical tests in REST 2008 now include correction for variation in efficiency. If variation in efficiency is low, hypothesis tests will produce more conclusive results, and confidence bands for estimated expression will be smaller. As all statistics are calculated using randomisation techniques, the approach for measuring standard curve error must also be stochastic, and is expressed as a challenge: If we ignore variation in thestandard curve, the slope (m value) will be expressed as a constant in all equations. Say, then, we have a standard curve of six data points for the gene GAPDH that we use to estimate its efficiency. If there is no variation in the standard curve, then we could pick any two points in the curve and still measure the same gradient. If, however, there is large variation between the points, then random selection of points will greatly vary the efficiency calculated. Using a few data points, we can then simulate the random variable representing the efficiency error. The randomised efficiency value is then included in calculations instead of the slope of the line of best fit, feeding any variation in efficiency directly into the relative quantitation hypothesis tests and confidence intervals.

Whisker-Box Plots

REST 2008 replaces the bar graph visualisation in prior versions with a statistical whisker-box plot. Instatistical applications, whisker-box plots provide additional information about the skew of distributions that would not be available simply by plotting the sample mean. See the link below for general information about whisker-box plots: http://regentsprep.org/Regents/math/data/boxwhisk.htm

References & Links

"Relative Expression Software Tool (REST) for group-wise comparison and statistical analysis of relative expression results in Real-Time PCR", (Pfaffl et al, 2002)

"Accurate normalization of real-time quantitative RT-PCR data by geometric averaging of multiple internal control genes" (Vandersompele et al, 2002)

"Bootstrap Methods and their Application" (A.C. Davidson, D.V. Hinkley 2002), Cambridge University Press (ISBN 0-521-57391-2, Cambridge University Press 2002)

Corbett Research Ptd Ltd: Rotor-Gene 6000 Series Software User Guide (Australia, Sydney, 2008) (Corbett Research)

This reference provides a good introduction to the philosophy of randomised tests:
http://ordination.okstate.edu/permute.htm

This reference provides an online interactive example of the test:
http://www.bioss.ac.uk/smart/unix/mrandt/slides/frames.htm

This reference provides more detailed descriptions on how to carry out traditional tests, such as determination of confidence intervals and hypothesis testing using bootstrapping and randomisation:
http://www.uvm.edu/~dhowell/StatPages/Resampling/Resampling.html

A description of Whisker-Box Plots:
http://regentsprep.org/Regents/math/data/boxwhisk.htm

Contact Information

If you have further questions or comments to improve the software, your suggestion are always welcome.
Please contact us at this address: rest-2008@gene-quantification.info?subject=REST-2008