[Genabel-commits] r1196 - pkg/OmicABEL/doc

[noreply at r-forge.r-project.org]

Author: dfabregat
Date: 2013-04-17 12:11:37 +0200 (Wed, 17 Apr 2013)
New Revision: 1196

Added:
   pkg/OmicABEL/doc/InternalOutputFormat.tex
Log:
[Documentation] Adding a file describing the format of the output files generated by CLAK-GWAS (*.out, *.iout).


Added: pkg/OmicABEL/doc/InternalOutputFormat.tex
===================================================================
--- pkg/OmicABEL/doc/InternalOutputFormat.tex	                        (rev 0)
+++ pkg/OmicABEL/doc/InternalOutputFormat.tex	2013-04-17 10:11:37 UTC (rev 1196)
@@ -0,0 +1,137 @@
+\documentclass[a4paper,10pt]{article}
+
+\usepackage{amssymb}
+\usepackage{amsmath}
+
+\addtolength{\textwidth}{2.2cm}  
+\addtolength{\oddsidemargin}{-1.10cm}
+\addtolength{\evensidemargin}{-1.10cm}
+\addtolength{\textheight}{2.0cm} 
+\addtolength{\topmargin}{-1.2cm}
+
+\usepackage{colortbl}
+
+\usepackage{fancybox}
+\usepackage{pgf}
+\usepackage{tikz}
+\usetikzlibrary{calc,shapes,snakes,arrows,shadows}
+
+
+\begin{document}
+
+\title{About the format of the B.out and B.iout files}
+
+\author{Diego Fabregat-Traver}
+
+\date{November, 2012}
+
+\maketitle
+
+The output from CLAK-GWAS is divided in two files: {\tt B.out} and {\tt B.iout}.
+The first stores the actual data, the second the metadata.
+
+\vspace{3mm}
+\noindent
+The metadata file format is as follows:
+
+\begin{verbatim}
+int type     // Double -> 6 as in Databel
+int nbytes   // Double -> 8
+int p        // # of covariates + 2 (intercept and SNP)
+int m        // # of SNPs
+int t        // # of traits
+int tile_m   // More on these two later
+int tile_t
+int namelength // 32 as in Databel
+char * labels  // More in a second
+\end{verbatim}
+
+\vspace{3mm}
+\noindent
+Labels:
+
+\begin{itemize}
+	\item Beta labels: {\tt beta\_intercept}, {\tt beta\_cov1}, ... , {\tt beta\_SNP}
+	\item SE labels: {\tt se\_intercept}, {\tt se\_cov1}, ... , {\tt se\_SNP}
+	\item Covariates labels: {\tt cov\_cov1\_intercept}, {\tt cov\_cov2\_intercept}, ... , {\tt cov\_SNP\_intercept}, {\tt cov\_cov2\_cov1}, ...
+\end{itemize}
+
+\vspace{3mm}
+\noindent
+Essentially, the labels are the betas, then the diagonal of the var-cov matrix, and then the lower triangle by columns.
+
+
+\vspace{3mm}
+\noindent
+The data file ({\tt B.out}) is simply a long sequence of doubles. 
+The way data is stored is a bit complicated if you are not used to this. Let me explain
+based on the figure below.
+The entire grid of results for $m$ SNPs $\times$ $t$ traits is stored in blocks or {\em tiles} of size
+$tile_m \times tile_t$. Tiles are stored column-wise: First {\tt T0}, then {\tt T1} and so on.
+Within each tile, data is also stored column-wise: (Trait0, SNP0), (Trait0, SNP1), (Trait0, SNP2), ..., (Trait0, SNP(m-1)), (Trait1, SNP0), ...
+
+	\begin{tikzpicture}
+
+		\draw [ystep=2.0,xstep=2.0] (0,0) grid (6,6);
+
+		\draw[black, latex-latex] ( 0, 7.0 ) -- ( 6, 7.0 );
+		\draw[black, latex-latex] ( -1.0, 0 ) -- ( -1.0, 6 );
+		\draw[black, latex-latex] ( 0, 6.4 ) -- ( 2, 6.4 );
+		\draw[black, latex-latex] ( -0.4, 4 ) -- ( -0.4, 6 );
+		\node[at={(3, 7.2)}] {$t$};
+		\node[at={(-1.2, 3)}] {$m$};
+		\node[at={(1, 6.6)}] {$tile_t$};
+		\node[at={(-0.4, 5)}] {$tile_m$};
+
+		\node[at={( 1, 5 )}] {\tt T0};
+		\node[at={( 1, 3 )}] {\tt T1};
+		\node[at={( 1, 1 )}] {\tt T2};
+		\node[at={( 3, 5 )}] {\tt T3};
+		\node[at={( 3, 3 )}] {\tt T4};
+		\node[at={( 3, 1 )}] {\tt T5};
+		\node[at={( 5, 5 )}] {\tt T6};
+		\node[at={( 5, 3 )}] {\tt T7};
+		\node[at={( 5, 1 )}] {\tt T8};
+		
+	\end{tikzpicture}
+
+\vspace{15mm}
+\noindent
+Now, each entry (Trait*, SNP*) is actually an array containing all the values for the 
+corresponding beta and lower triangle of the var-cov. For instance, if we have 2 covariates
+(sex, age) and a SNP (rs33), with a var-cov matrix:
+
+\vspace{5mm}
+\begin{tabular}{| c | c  c  c  c |} \hline
+	\# & Int & sex & age & rs33 \\\hline
+	Int &  se\_Int & * & * & * \\
+	sex &  cov\_sex\_Int & se\_sex & * & * \\
+	age &  cov\_age\_Int & cov\_age\_sex & se\_age & * \\
+	rs33 & cov\_rs33\_Int & cov\_rs33\_sex & cov\_rs33\_age & se\_rs33 \\\hline
+\end{tabular}
+
+
+\vspace{5mm}
+\noindent
+The array is as follows:
+
+\begin{verbatim}
+[beta_Int, beta_sex, beta_age, beta_rs33, se_Int, se_sex, se_age, se_rs33, 
+ cov_sex_Int, cov_age_Int, cov_rs33_Int, cov_age_sex, cov_rs33_sex, cov_rs33_age]
+\end{verbatim}
+
+\vspace{1mm}
+\noindent
+Notice that the order matches the one for the labels in the file {\tt B.iout}.
+
+
+\vspace{5mm}
+\noindent
+All the data above occupies $m \times t \times (p + p*(p+1)/2)$ doubles. After all these values,
+you can find the estimates: heritability, sigma, res\_sigma, and beta estimates, 
+for a total of $(3 + (p-1)) \times t$ values. The storage is the following: 
+first the $t$ heritability values (again trait0, trait1, ..., trait(t-1));
+then the $t$ sigmas; then $t$ res\_sigmas; and finally the $t$ beta estimates.
+The beta estimates are stored as expected: beta\_Int, beta\_cov1, ..., beta\_cov(p-2).
+
+\end{document}