[R-gregmisc-users] SF.net SVN: r-gregmisc: [970] trunk/ssize/inst/doc/ssize.Rnw
Brought to you by:
warnes
Menu
▾
▴
[R-gregmisc-users] SF.net SVN: r-gregmisc: [970] trunk/ssize/inst/doc/ssize.Rnw
|
From: <wa...@us...> - 2006-07-24 21:23:39
|
Revision: 970 Author: warnes Date: 2006-07-24 14:23:26 -0700 (Mon, 24 Jul 2006) ViewCVS: http://svn.sourceforge.net/r-gregmisc/?rev=970&view=rev Log Message: ----------- Put into RNews format. Modified Paths: -------------- trunk/ssize/inst/doc/ssize.Rnw Modified: trunk/ssize/inst/doc/ssize.Rnw =================================================================== --- trunk/ssize/inst/doc/ssize.Rnw 2006-06-29 18:11:22 UTC (rev 969) +++ trunk/ssize/inst/doc/ssize.Rnw 2006-07-24 21:23:26 UTC (rev 970) @@ -4,29 +4,24 @@ %\VignettePackage{ssize} -\documentclass[12pt]{article} -\usepackage{url} -\usepackage{amsmath} -\usepackage{natbib} +%\documentclass[letter]{report} +\documentclass[a4paper]{report} +\usepackage{/Library/Frameworks/R.framework/Resources/share/texmf/Sweave} +\usepackage{Rnews} +\usepackage[round]{natbib} - -\newcommand{\Robject}[1]{{\texttt{#1}}} -\newcommand{\Rfunction}[1]{{\texttt{#1}}} -\newcommand{\Rpackage}[1]{{\textit{#1}}} -\newcommand{\Rclass}[1]{{\textit{#1}}} -\newcommand{\Rmethod}[1]{{\textit{#1}}} -\newcommand{\code}[1]{\texttt{#1}} - \begin{document} +\begin{article} \title{Sample Size Estimation for Microarray Experiments Using the - \code{ssize} package.} + \code{ssize} package.} \author{Gregory R. Warnes \\ - email:\code{gre...@pf...}} + email:\code{wa...@bs...}} +\subtitle{ ~ } \maketitle -\begin{abstract} +\section*{abstract} RNA Expression Microarray technology is widely applied in biomedical and pharmaceutical research. The huge number of RNA concentrations @@ -39,20 +34,19 @@ the Bioconductor project (\url{http://www.bioconductor.org}) web site. -\end{abstract} +\section*{Note} -\section{Note} - This document is a simplified version of the manuscript \begin{quote} - Warnes, G. R., Liu, P. (2005) - Sample Size Estimation for Microarray Experiments, \emph{submitted - to} {\it Biometrics}. + Warnes, G. R., Liu, P. (2006) Sample Size Estimation for Microarray + Experiments, Technical Report, Department of Biostatisticsa and + Computational Biology, University of Rochester. \end{quote} -Please refer to that document for a detailed discussion of the -sample size estimation method. +which has been available as a pre-publication manuscript since 2004. +Please refer to that document for a detailed discussion of the sample +size estimation method. -\section{Introduction} +\section*{Introduction} High-throughput microarray experiments allow the measurement of expression levels for tens of thousands of genes simultaneously. @@ -72,9 +66,9 @@ essential to take into account multiple testing and dependency among variables when calculating sample size. -\section{Method} +\section*{Method} -\subsection{Overview} +\subsection*{Overview} \citet{Warnes05} provides a simple method for computing sample size for micrarray experiments, and reports on a sereies of simulations @@ -89,7 +83,7 @@ exceptionally valuable in helping scientific clients to make the difficult trade offs between experiment cost and statistical power. -\subsection{Assumptions} +\subsection*{Assumptions} In the current implementation, we assume that a microarray experiment is set up to compare gene expressions between one @@ -112,7 +106,7 @@ where $\mu_{T}$ and $\mu_{C}$ are means of gene expressions for treatment and control group respectively. -\subsection{Computations} +\subsection*{Computations} The proposed procedure to estimate sample size is: @@ -178,7 +172,7 @@ \ref{fig:CumFoldChangePlot}. -\section{Example} +\section*{Example} First, we need to load the \code{ssize} library: @@ -186,6 +180,7 @@ library(ssize) library(xtable) library(gdata) # for nobs() +options(width=30) @ As part of the \code{ssize} library, I've provided an example data @@ -196,39 +191,41 @@ <<>>= data(exp.sd) -str(exp.sd) +#str(exp.sd) @ This data was calculated via something like \begin{verbatim} library(affy) -setwd("/data/rstat-data/Standard_Affymetrix_Analysis/GT_methods_050316/WORK") +setwd("~/GT_methods_050316/WORK") load("probeset_data.Rda") expression.values <- exprs(probeset.data) covariate.data <- pData(probeset.data) -controls <- expression.values[,covariate.data$GROUP=="Control"] #$ +controls <- expression.values[, + covariate.data$GROUP=="Control"] #$ exp.sd <- apply(controls, 1, sd) \end{verbatim} Lets see what the distribution looks like: -\begin{figure}[h!] - \centering - \caption{Distribution of exp.sd} - \label{exp.sd.hist} -<<fig=TRUE,width=12,height=6>>= - hist(exp.sd,n=20, col="cyan", border="blue", main="", - xlab="Standard Deviation (for data on the log scale)") +%\begin{figure}[h!] +% \centering +% \caption{Distribution of exp.sd} +% \label{exp.sd.hist} +%\end{figure} + +<<fig=TRUE,width=12,height=12>>= + xlab <- c("Standard Deviation", " (for data on the log scale)") + hist(exp.sd,n=20, col="cyan", border="blue", main="", xlab=xlab) dens <- density(exp.sd) - lines(dens$x, dens$y*par("usr")[4]/max(dens$y),col="red",lwd=2) #$ - title("Histogram of Standard Deviations (log2 scale)") + scaled.y <- dens$y*par("usr")[4]/max(dens$y) + lines(dens$x,scaled.y ,col="red",lwd=2) #$ + title("Histogram of Standard Deviations") @ -\end{figure} -\begin{center} -\end{center} + Note that this distribution is right skewed, even though it is on the $\log_2$ scale. @@ -241,24 +238,29 @@ @ There are 6 functions available in the \code{ssize} package. -<<eval=FALSE>>= -?pow -@ \begin{verbatim} - pow(sd, n, delta, sig.level, alpha.correct = "Bonferonni") - power.plot(x, xlab = "Power", ylab = "Proportion of Genes with Power >= x", - marks = c(0.7, 0.8, 0.9), ...) + pow(sd, n, delta, sig.level, + alpha.correct = "Bonferonni") + power.plot(x, xlab = "Power", + ylab = "Proportion of Genes with" + " Power >= x", + marks = c(0.7, 0.8, 0.9), ...) - ssize(sd, delta, sig.level, power, alpha.correct = "Bonferonni") - ssize.plot(x, xlab = "Sample Size (per group)", - ylab = "Proportion of Genes Needing Sample Size <= n", - marks = c(2, 3, 4, 5, 6, 8, 10, 20), ...) + ssize(sd, delta, sig.level, power, + alpha.correct = "Bonferonni") + ssize.plot(x, + xlab = "Sample Size (per group)", + ylab = "Proportion of Genes Needing Sample" + " Size <= n", + marks = c(2, 3, 4, 5, 6, 8, 10, 20), ...) - delta(sd, n, power, sig.level, alpha.correct = "Bonferonni") - delta.plot (x, xlab = "Fold Change", - ylab = "Proportion of Genes with Power >= 80\% at Fold Change=delta", - marks = c(1.5, 2, 2.5, 3, 4, 6, 10), ...) + delta(sd, n, power, sig.level, + alpha.correct = "Bonferonni") + delta.plot (x, xlab = "Fold Change", + ylab = "Proportion of Genes with " + "Power >= 80\% at Fold Change=delta", + marks = c(1.5, 2, 2.5, 3, 4, 6, 10), ...) \end{verbatim} You will note that there are three pairs. @@ -297,10 +299,7 @@ \item What is the power for 6 patients per group with $\delta=1.0$, $\alpha=0.05$? -\begin{figure}[h!] - \caption{Effect of Sample Size on Power} \label{fig:CumNPlot} - \centering -<<fig=TRUE,width=12,height=6>>= +<<fig=TRUE,width=12,height=12>>= all.power <- pow(sd=exp.sd, n=n, delta=log2(fold.change), sig.level=sig.level) @@ -314,18 +313,16 @@ xjust=1, yjust=1, cex=1.0) title("Power to Detect 2-Fold Change") @ -\end{figure} +%\begin{figure}[h!] +% \caption{Effect of Sample Size on Power} \label{fig:CumNPlot} +% \centering +%\end{figure} - \item What is the necessary per-group sample size for $80\%$ power when $\delta=1.0$, and $\alpha=0.05$? -\begin{figure}[h!] - \caption{Sample size required to detect a 2-fold treatment effect.} - \label{fig:CumPowerPlot} - \centering -<<fig=TRUE,width=12,height=6>>= +<<fig=TRUE,width=12,height=12>>= all.size <- ssize(sd=exp.sd, delta=log2(fold.change), sig.level=sig.level, power=power) ssize.plot(all.size, lwd=2, col="magenta", xlim=c(1,20)) @@ -339,21 +336,27 @@ xjust=1, yjust=0, cex=1.0) title("Sample Size to Detect 2-Fold Change") @ -\end{figure} +%\begin{figure}[h!] +% \caption{Sample size required to detect a 2-fold treatment effect.} +% \label{fig:CumPowerPlot} +% \centering +%\end{figure} -\clearpage +%\clearpage \item What is necessary fold change to achieve $80\%$ with $n=6$ patients per group, when $\delta=1.0$ and $\alpha=0.05$? -\begin{figure}[h!] - \caption[Given Sample Size, Fold Change (Effect Size) Necessary to - Achieving a Specified Power]{Given sample size, this plot allows - visualization of the fraction of genes achieving the specified - power for different fold changes.} - \label{fig:CumFoldChangePlot} - \centering -<<fig=TRUE,width=12,height=6>>= +%\begin{figure}[h!] +% \caption[Given Sample Size, Fold Change (Effect Size) Necessary to +% Achieving a Specified Power]{Given sample size, this plot allows +% visualization of the fraction of genes achieving the specified +% power for different fold changes.} +% \label{fig:CumFoldChangePlot} +% \centering +%\end{figure} + +<<fig=TRUE,width=12,height=12>>= all.delta <- delta(sd=exp.sd, power=power, n=n, sig.level=sig.level) delta.plot(all.delta, lwd=2, col="magenta", xlim=c(1,10)) @@ -366,11 +369,10 @@ xjust=1, yjust=0, cex=1.0) title("Fold Change to Achieve 80\% Power") @ -\end{figure} \end{enumerate} -\section{Modifications} +\section*{Modifications} While the \code{ssize} package has been implemented using the simple 2-sample pooled t-test, you can easily modify the code for other @@ -379,13 +381,13 @@ appropriate computation for the desired experimental design. -\section{Future Work} +\section*{Future Work} Peng Liu is currently developing methods and code for substituting False Discovery Rate for the Bonferonni multiple comparison adjustment. -\section{Contributions} +\section*{Contributions} Contributions and discussion are welcome. @@ -452,4 +454,6 @@ 653-667. \end{thebibliography} + +\end{article} \end{document} This was sent by the SourceForge.net collaborative development platform, the world's largest Open Source development site. |
