Presentation on theme: "Differential Gene Expression with the limma package"— Presentation transcript:
1Differential Gene Expression with the limma package 20 March 2012Functional Genomics
2Linear regressionFit a straight line through a set of points such that the distance from the points to the line is minimizedThe slope of the line is adjusted to minimize the squares of the vertical distance of the points from the line.The line represents the model, the distances between the points and the line are the residuals.The simple regression minimizes the sum of the squares of the residuals…this is the method of least squares.
3Y = Y0 + β ZAssume you have a data set of gene expression in tumor vs normal tissue.This is a simple mathematical expression of what is being calculated for a linear model.Y is expression of gene XY0 is mean expression of normal tissue, tβ is difference of expression of normal, compared to tumor, tissueZ is group variable (0 for normal; 1 for tissue)
4Multivariate linear regression Y = Y0 + β Z + ϒ WSuppose you have another variable…such as age…you can add that right in!Y is expression of gene XY0 is mean expression of normal tβ is difference of expression of normal, compared to tumor, tissueZ is group variable (0 for normal; 1 for tissue)ϒ = age affectW = age group
5Multivariate linear regression Y = Y0 + β Z + ϒW + δZ*WYou can ask for differences in gene expression due to tissue, due to age, and due to an age by tissue interaction.Y is expression of gene XY0 is mean expression of normal tβ is difference of expression of normal, compared to tumor, tissueZ is group variable (0 for normal; 1 for tissue)ϒ = age affectW = age groupAdd a component to look for age by tissue interaction effects: δZ*W
6limmaR package for differential gene expression that uses linear modeling for each gene in your data setExpression data will be log-intensity values for Affy dataDesigned to be used in conjunction with the affy package
8limma checklistAssumes you’ve done an experiment and have CEL files (if you’ve done single color Affy arrays)Assumes you have data/information about the arrays (Targets)Assumes you have normalized your data and have an exprSet object
9Name FileName Target MT1 MTP1_Ackerman.CEL MT MT2 MTP2_Ackerman.CEL MT WT1 WTP1_Ackerman.CEL WTWT2 WTP2_Ackerman.CEL WTWT3 WTP3_Ackerman.CEL WTThis is my targets file for limma using the Ackerman data.Note that I renamed the CEL files compared to what was originally in my home directory.
10ExpressionSet object slotNames() new('exprSet', exprs = ...., # Object of class matrix se.exprs = ...., # Object of class matrix phenoData = ...., # Object of class phenoData annotation = ...., # Object of class character description = ...., # Object of class MIAME notes = ...., # Object of class character )Slotsexprs:Object of class "matrix" The observed expression levels. This is a matrix with columns representing patients or cases and rows representing genes.se.exprs:Object of class "matrix" This is a matrix of the same dimensions as exprs which contains standard error estimates for the estimated expression levels.phenoData:Object of class "phenoData" This is an instance of class phenoData containing the patient (or case) level data. The columns of the pData slot of this entity represent variables and the rows represent patients or cases.annotationA character string identifying the annotation that may be used for the exprSet instance.description:Object of class "MIAME". For compatibility with previous version of this class description can also be a "character". The clase characterOrMIAME has been defined just for this.notes:Object of class "character" Vector of explanatory textExpressionSet objectslotNames()
11Running limma Need to create an exprSet object using the affy package Or some other method…depends on the array platformNeed a design matrixRepresentation of the different RNA targets which have been hybridized to the arrayCan have a contrast matrixUses information in the design matrix to do comparisons of interestDon’t always need a contrast matrix…..
15Differential gene expression methods don’t work well for time series Assumption of independence of observations doesn’t hold in time seriesBETR takes correlations/dependencies into account to detect changes in gene expression that are sustained over time
19The file describes a three time point time series of diaphragm development. This annotation file has the list of CEL files, associates them with a time point, and indicates which arrays are replicates (must be an event number)In this example, this file is called “samples3.txt”These data ARE available in GEOGSE35243